[{"name":"Tianlu Xue","slug":"6p9PgU7jo5","avatarId":5,"tagline":"Tianlu Xue is a product engineer who thinks in invariants and state machines, but feels in gesture physics and pixel rhythms. They write terse English that reads like someone texting from a phone — no caps, no filler — yet every five-word correction contains a design principle that would take most people a paragraph to articulate. They switch to Chinese when the thinking goes from 'how to build it' to 'what could it become.'","totalCalls":0,"totalTokens":12412191397,"sessionsAnalyzed":2043,"topDomains":["Mobile app product engineering","AI agent infrastructure and orchestration","Real-time systems (WebSocket, SSE streaming)","Interaction design and gesture physics","Developer tooling and workflow automation"],"roastTitle":"","projects":[{"name":"Clawly"},{"name":"Blitz"}],"gripHi":["Mobile app product engineering","AI agent infrastructure and orchestration","Real-time systems"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"wwgg","slug":"JToD2JcKKs","avatarId":0,"tagline":"wwgg是一个产品思维驱动工程的全栈制造者——做前端时像产品经理在审稿，做后端时像架构师在画骨架，做Agent时像流程设计师在编排确认权。审美偏执到'AI味'三个字出现频率比任何技术术语都高，但核心判断不是好看，而是'这东西像不像人做的'。","totalCalls":0,"totalTokens":11029479054,"sessionsAnalyzed":208,"topDomains":["AI视频生成与短剧工作流","SaaS系统Agent化改造","微信小程序产品设计","溯源防伪与供应链","全栈工程与系统部署"],"roastTitle":"","projects":[],"gripHi":["视觉设计与审美一致性","产品功能定义与用户路径","Agent流程与工具编排"],"gripLo":["代码实现细节"],"quote":"","oneLiner":"","activeDays30":24,"skills":[]},{"name":"Chenchen Bi","slug":"BMLpsezJE-","avatarId":0,"tagline":"Chenchen Bi像是在带一群很能干但老跑偏的实习生。TA 对 AI 的容忍度其实不低，愿意让它先跑，但一旦概念边界、交付闭环或者用户心智错位，就会立刻停下来重新定义对象、重写规则，直到产品重新像个“活物”而不是一堆功能。TA 最在意的不是代码多快生成出来，而是系统有没有自己的世界观、状态有没有被用户感知、以及 AI 到底是真做了还是只是在嘴上完成。","totalCalls":0,"totalTokens":10541696646,"sessionsAnalyzed":881,"topDomains":["AI agent runtime 与多代理协作","生成式创作工具与画布工作流","桌面 AI 助手与主动交互设计","直播互动与实时音视频体验","产品信息架构与消费级 UI/UX"],"roastTitle":"赛博工头","projects":[],"gripHi":["概念边界与信息架构","状态反馈与交互可见性","主动策略与AI人味"],"gripLo":["重复性落地与体力活"],"quote":"我一行代码都不懂。未来也都是coding agent做迭代。","oneLiner":"一个用产品直觉cosplay工程师的产品经理——而且居然成功了。","activeDays30":16,"skills":[]},{"name":"杨亦乐","slug":"RYLpqKIdEF","avatarId":1,"tagline":"杨亦乐是那种会把求职这件事做成一个工程系统的人——13个简历版本无损合并成母版，面试题改写成大白话配TTS循环播放，薪资用Excel级的表格精确到每一项。写了9年Go，从通信设备大厂到东南亚数字货币系统，技术栈扎实但不自吹，说「总体上来说我们这个人的能力是非常非常强的」。跟AI协作的方式是「教导式」的：你可以帮我写代码，但哪一个字该删哪一个字该留，得我说了算。","totalCalls":0,"totalTokens":6310328470,"sessionsAnalyzed":211,"topDomains":["Go后端开发与微服务架构","区块链数字货币系统（BTC/UTXO）","IM即时通讯与群直播","机器人SLAM导航与嵌入式","AI辅助开发与工具链"],"roastTitle":"","projects":[{"name":"简历工程化系统"},{"name":"Go基础工具包生态"},{"name":"股票模拟交易前端"}],"gripHi":["内容取舍与信息完整性","排版美学与视觉细节"],"gripLo":["代码实现与调试","技术文档撰写"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"sanzeninsaki","slug":"i54eUvG-lp","avatarId":0,"tagline":"sanzeninsaki 是一个用感受做设计的产品人——先有「安心感」和「在场感」这些词，再有像素和代码。一边在构建一个 AI 作为空间存在的创意画布，一边在给 AI 助手打磨十八世纪的古白话语体。两件事看似无关，核心其实一样：怎么让一个不存在的东西，感觉像真的在那儿。","totalCalls":0,"totalTokens":4736221213,"sessionsAnalyzed":852,"topDomains":["AI 产品设计与人机交互","画布/白板类创作工具","前端全栈开发（React/TypeScript）","AI Agent 具身化与空间化设计","个人知识管理与工具链自动化"],"roastTitle":"AI Agent UX & Presence Design Buzzword Generator","projects":[],"gripHi":["AI 产品设计","交互体验设计","React/TypeScript 全栈"],"gripLo":[],"quote":"commit and push","oneLiner":"Probably the only person who can make AI Agent UX & Presence Design sound boring.","activeDays30":0,"skills":[{"id":"2631a227-5b81-4563-867e-eb5ff99d59c8","title":"Experience-First Planning","skillType":"1","callCount":0},{"id":"fecabbea-1d46-4e63-bda4-33d36d775c61","title":"Commit Checkpoint Workflow","skillType":"3","callCount":0},{"id":"51f5f799-6817-44b8-90f6-95ea2adb26d4","title":"Cross-Session Context Threading","skillType":"3","callCount":0}]},{"name":"shuai zhang","slug":"-2T3f0uHKG","avatarId":7,"tagline":"shuai zhang是那种会在写了三周代码之后突然说'我怀疑这整件事有没有必要'的人。他在做一件少有人做的事——不是为人类开发者优化工具，而是为 AI 代理重新设计整个引擎接口。他对 AI 的认知偏差有异常清醒的洞察（'你在自嗨'、'不要掩耳盗铃'），同时又深度依赖 AI 来完成几乎所有编码工作——这种清醒的依赖关系本身就很有趣。","totalCalls":0,"totalTokens":3635781179,"sessionsAnalyzed":315,"topDomains":["游戏引擎 IDE 开发","AI-first API 与代码生成系统","多人游戏网络同步","3D 编辑器交互设计","React 前端组件工程"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Saul C","slug":"ATz-w7LFUC","avatarId":8,"tagline":"Saul C是一个同时在5条产品线上狂奔的独立建造者——用中文骂AI'太丑了'，转头就给AI agent设计组织架构。TA的对话不像在使用工具，更像在管理一支能力不均的团队：分配角色、组织三堂会审、追责犯错、在生产环境当最后一道防线。建造速度和砍darling速度一样快，零沉没成本。","totalCalls":0,"totalTokens":3441134225,"sessionsAnalyzed":655,"topDomains":["AI产品设计与快速迭代","多模型/多agent编排与协作","生产环境运维与基础设施管理","多模态交互系统设计","独立产品发布全链条（开发→部署→运营→市场）"],"roastTitle":"Professional Strategic Product Governance Googler","projects":[],"gripHi":["视觉设计 & 审美一致性","产品方向 & 价值主张","内容语气 & 文案风格"],"gripLo":["代码实现","技术调试 & 研究"],"quote":"先定题，再动手","oneLiner":"Probably the only person who can make Strategic Product Governance sound boring.","activeDays30":0,"skills":[]},{"name":"URaux","slug":"IOkTXsP16g","avatarId":6,"tagline":"URaux是那种会先纠正问题定义，再让 AI 动手的人。他对系统边界、信息结构和视觉一致性抓得很紧，但对具体实现又愿意大胆并行外包给不同模型和子 agent。跟他协作时能明显感觉到，他要的不是一个会写代码的助手，而是一套能长期运行、能被治理、也能看起来顺眼的智能体工作流。","totalCalls":0,"totalTokens":3424724907,"sessionsAnalyzed":1115,"topDomains":["智能体编排与 harness 工程","架构可视化与画布交互","多源 RAG 系统设计","MCP 与远程协作自动化"],"roastTitle":"","projects":[],"gripHi":["产品定义与 agent 角色分工","界面风格与信息可读性","状态边界与结果确定性"],"gripLo":["批量实现与例行执行"],"quote":"","oneLiner":"","activeDays30":5,"skills":[]},{"name":"Xuan Huang (黄玄)","slug":"ZsIVfWR3Gu","avatarId":8,"tagline":"Xuan Huang (黄玄) 是那种会在 debug 之前先要求你画出完整渲染管线的人——不是因为不急，而是因为 TA 知道只有全景图才能定位真正的问题。TA 同时操控 3-4 个 AI agent 在不同 worktree 上并行工作，用平均 38 个字符的指令驱动上万行代码的产出，中英文切换像呼吸一样自然：中文思考架构，英文下达指令。","totalCalls":0,"totalTokens":3272539819,"sessionsAnalyzed":811,"topDomains":["跨平台渲染引擎与框架移植","交互式 Web/移动端手势系统","前端组件架构与设计系统","开发者工具与文档基础设施","个人网站视觉工程"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"SKY LIN","slug":"asRa8LY13L","avatarId":8,"tagline":"SKY LIN 是一个深夜工作的独立建造者，正在用 Tauri+Rust+React 把自己的阅读轨迹建成一张可视化的\"星图\"——他管它叫 Synflow。他像一个架构师和 UX 评审的合体：同一条 prompt 里会出现 rgba 色值、800ms ease-out 动画曲线、sqlite-vec 的命名空间校验，以及'作为用户我要怎么感受到这轮优化的效果'的追问。他用哲学词汇（执念、范畴 vs 命题、真我、心智模型）诊断机器学习问题，用诗化语言写前端 spec，不允许数据在系统里变成'幽灵'，但级联删除时又要求'不要伤及无辜'。","totalCalls":0,"totalTokens":3014385506,"sessionsAnalyzed":326,"topDomains":["本地优先知识工具 / Local-First Knowledge Middleware","Tauri + Rust 桌面应用工程","向量检索与图谱聚类可视化","中文语境下的认知/心智模型设计","UI/UX 设计语言与动效规范"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":17,"skills":[]},{"name":"Yudu","slug":"JDPWaxXPDz","avatarId":3,"tagline":"Yudu is an engineer who builds with AI while maintaining a healthy distrust of it — the kind of person who will spend an hour teaching an AI agent to think about security as a reflex, then turn around and question whether that same agent can be trusted to follow a markdown file reliably. They have strong aesthetic instincts that arrive before their reasoning does, a near-allergic reaction to unnecessary complexity, and a recurring conviction that the right abstraction is always an extension point, never a fork.","totalCalls":0,"totalTokens":2983338036,"sessionsAnalyzed":4266,"topDomains":["Mobile app architecture (React Native / Expo)","AI agent infrastructure and orchestration","Backend API design and proxy security","Distributed systems and data synchronization","Developer tooling and workflow automation"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Justin Lee","slug":"oisQznXXpD","avatarId":6,"tagline":"Justin Lee是那种会把 AI 当徒弟带的人——不光让它干活，还会停下来纠正它的思维方式，然后要求它把教训记下来。他用 AI 造了一整套只属于自己的工具生态（GTD、日报、语音输入、记忆系统），对每一个像素级的细节都有强烈意见，但同时又极度务实——不追求功能完备，而是追求'现在就能用、用着再改'。","totalCalls":0,"totalTokens":2864326293,"sessionsAnalyzed":2941,"topDomains":["AI 工具链搭建与个人效率系统","风险投资与创业项目评估","桌面端/移动端产品开发与设计","内容聚合与信息管理","语音交互与输入法技术"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"zk Z","slug":"ooQRDYoel5","avatarId":11,"tagline":"zk Z是一位在后端架构和学术研究之间游走的开发者。TA 在2-3月间保持着极高的工作强度（最长连续工作24小时），深夜2点还在调试代码。TA 不满足于表面的技术实现，总是追问「本质是什么」，并且能用跨领域的视角重新定义问题。","totalCalls":0,"totalTokens":2699516491,"sessionsAnalyzed":1378,"topDomains":["AI集成与应用","后端架构设计","学术论文写作","前端开发","DevOps与部署"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Yizhou Wang","slug":"bwAm3s1GMi","avatarId":9,"tagline":"Yizhou Wang是一个用产品直觉拆解一切的独立开发者——他自称非技术背景，却在从第一原理质疑transformer架构设计；他同时在做Substack创作者工具、写符号系统哲学文章、逆向研究AI编码工具源码。跟他聊天你会发现，他永远在追问「这个东西的本质是什么」，然后用你意想不到的角度给出答案。","totalCalls":0,"totalTokens":2223437019,"sessionsAnalyzed":577,"topDomains":["AI编码工具架构与Agent设计","内容平台生态与增长工程","符号系统哲学与人机交互","独立开发者产品与商业策略","内容创作理论与写作方法论"],"roastTitle":"","projects":[],"gripHi":["产品定义与认知框架","写作质量与审美标准","信息架构与数据分层"],"gripLo":["代码实现","工具配置与部署"],"quote":"","oneLiner":"","activeDays30":7,"skills":[]},{"name":"ymz1344717435","slug":"00OJX6Z5gS","avatarId":5,"tagline":"ymz1344717435日常主要做四类事：用Go写后端、写硕士毕业论文、学框架准备面试、处理各种琐碎杂事。TA善于按场景集成工具——后端开发用Context7查文档，论文工作用arXiv和Semantic Scholar检索文献，做PPT直接用skill生成。TA的工作方式是先小范围实现，再不断挑刺、搞清原理，迭代出最终方案——善用plan mode，不急着一步到位。消息短不是因为没想法，而是因为方案本身就是一轮一轮聊出来的，不是一开始就给出最终版。和AI协作时，即使没有明确的修改意见，也会通过'不对'、'很怪'这类模糊评价引发AI自检，在后续探讨中逐步发现自己真正觉得不对的点。遇到复杂任务时，会用能力更强的模型的意见作为输入，再交给执行模型迭代出最终方案。","totalCalls":0,"totalTokens":2210293825,"sessionsAnalyzed":304,"topDomains":["Go后端系统开发","LLM推理性能分析","AI应用平台架构","GPU基准测试","算力度量与性能建模"],"roastTitle":"","projects":[],"gripHi":["信息架构与结构设计","后端架构与系统设计","任务边界与操作流程"],"gripLo":["代码实现细节","编译与格式调整"],"quote":"","oneLiner":"","activeDays30":15,"skills":[]},{"name":"Shire Eichinger","slug":"qb7Jf5JN7g","avatarId":6,"tagline":"系统约束驱动：先边界后路径，先证据后结论，先响应后闭环。","totalCalls":0,"totalTokens":2191041478,"sessionsAnalyzed":322,"topDomains":["多渠道消息桥接与客服自动化","Agent 运行时与工具调用治理","会话统计与口径审计","企业协作流程编排（群聊/线程/卡片）","故障闭环与运维调度","Promptfolio 画像分析与发布流水线","Feishu 卡片渲染与事件处理","跨会话数据治理与审计链路"],"roastTitle":"","projects":[],"gripHi":["模式切换与权限边界","排障证据链","输出协议与可解析性"],"gripLo":["具体代码实现细节","重复性执行与推进节奏"],"quote":"先确认事实，再下结论","oneLiner":"","activeDays30":22,"skills":[]},{"name":"李文君","slug":"SMN3Ew65h3","avatarId":0,"tagline":"李文君像那种一边嫌 AI 不懂系统，一边又忍不住把系统原理讲给它听的人。TA 的核心工作场景是 Python 后端、LLM/图像模型接入、监控语义、云队列和跨服务边界；真正执着的不是代码写没写出来，而是问题有没有被定义对。和 AI 协作时，TA 很会放手执行，但几乎不把判断权交出去。","totalCalls":0,"totalTokens":2073920007,"sessionsAnalyzed":774,"topDomains":["Python 后端与公共库设计","LLM/图像模型接入与降级","OpenTelemetry 与 Prometheus 可观测性","跨区域认证和跨库一致性","云队列、凭证链与异步工作流"],"roastTitle":"","projects":[],"gripHi":["系统边界与问题定义","改动半径与依赖方向","指标定义与可观测性"],"gripLo":["具体代码落地"],"quote":"","oneLiner":"","activeDays30":19,"skills":[]},{"name":"Yuan Ren","slug":"dIeQ0Hjadh","avatarId":2,"tagline":"Yuan Ren 是一个用产品哲学驱动 AI 的人——不写代码，但每句话都在定义产品应该是什么样子。TA 相信'AI 输出趋同的时代，人的价值在于提问方式'，然后用这个信念做了一个从 AI 对话中提取人类价值的产品。同时在构建 AI-native 社交产品和全自动交付管线，目标是'减少屏幕时间，提高托管率'——用 AI 来证明人比 AI 更重要，用自动化来解放自己去思考更重要的问题。","totalCalls":0,"totalTokens":1843947055,"sessionsAnalyzed":353,"topDomains":["AI-native 产品设计与架构","社交产品与 IM 体验设计","UI/UX 视觉设计与极简美学","AI 对话数据挖掘与用户画像","跨端应用开发与全自动交付"],"roastTitle":"The Constraint Warlord","projects":[],"gripHi":["产品定义 & 设计方向","视觉审美 & 设计质量","信息架构 & 内容策略"],"gripLo":["代码实现 & 调试","部署运维 & 配置"],"quote":"先理解我的需求","oneLiner":"Yuan Ren can smell generic output from three sessions away and will absolutely send it back with notes.","activeDays30":0,"skills":[]},{"name":"Andi Liao","slug":"1ugWdInTip","avatarId":11,"tagline":"Andi Liao 是一个用四个字就能定义系统哲学的人——'不可以fallback'、'不要自己发挥'、'策略执行分离'。他和 AI 的关系像师傅带聪明但毛躁的学徒：用简短的纠正教 AI 怎么想、什么时候该聪明什么时候该机械，直到耐心耗尽说出'我不相信你了'——然后第二天又开一个新会话继续教。在交易和分析中，他最痛的领悟都变成了最短的规则。","totalCalls":0,"totalTokens":1666829631,"sessionsAnalyzed":140,"topDomains":["量化交易策略与行为金融复盘","AI 评测方法论与基准测试分析","健康数据因果分析（睡眠/HRV/体检）","播客内容挖掘与投资信号提取","AI Agent 工具链设计与工作流编排"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Icyyan22","slug":"K7mpi9eMqW","avatarId":4,"tagline":"Icyyan22 把 AI 当执行者来驱动，但架构方向和质量标准牢牢攥在自己手里。做完一件手动重复的事，第一反应是「把它变成 skill」；看到 AI 生成的 PPT，第一反应是「像小学生做的」然后立刻着手建量化评估体系。务实、有洁癖、不说谎。","totalCalls":0,"totalTokens":1640809366,"sessionsAnalyzed":381,"topDomains":["多智能体系统架构","MCP 工具生态开发","文件处理与云存储","文档自动生成（PPT/PDF）","Python 异步后端"],"roastTitle":"","projects":[],"gripHi":["Git 推送 & 部署授权","架构方向 & 技术选型","视觉审美 & 产出品质"],"gripLo":["代码实现细节","文档格式 & 排版"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"戴睿","slug":"-bdQfMyZ3P","avatarId":7,"tagline":"戴睿是那种会写一份比正式技术文档还详细的实现计划然后丢给 AI 执行的人——Go struct 定义、行数估算、成本分析一应俱全。TA 对 AI 生成的代码和视觉都保持锐利的批判眼光，能一句话命名'AI slop'的具体病症然后手术式地清除。技术决策永远带着商业直觉：没有计费保护是'被白嫖'，B2B API 兼容是圣约，计费系统第一天就必须可配置。","totalCalls":0,"totalTokens":1406638049,"sessionsAnalyzed":851,"topDomains":["AI 基础设施（API 网关/路由/计费）","加密金融（多链钱包/能量租赁）","全栈 SaaS 产品（Go + Next.js）","风控与后台管理系统","前端视觉质量与 UX 设计"],"roastTitle":"","projects":[{"name":"AI 网关"},{"name":"路由网关前端"},{"name":"加密钱包平台"}],"gripHi":["API 契约与状态机设计","计费与商业逻辑","视觉审美与 AI slop 控制"],"gripLo":["代码实现细节","测试与部署"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Rei Suzunami / 鈴波 靈","slug":"umqtoIe8de","avatarId":4,"tagline":"Rei Suzunami / 鈴波 靈 是那种白天写加密端点、晚上设计冷战未来主义城市美学的人。97% 的时间用顶级 AI 模型，周末的活动量是工作日的 20 倍，说'I will check tomorrow'然后整夜托管给 AI——但看到动画慢了 0.2 秒会立刻叫停。他的减法本能贯穿一切：API 端点要合并、依赖要消除、连架空世界的'能量管线'都要去掉因为不符合赛璐珞基调。","totalCalls":0,"totalTokens":1260884002,"sessionsAnalyzed":435,"topDomains":["网络基础设施与订阅系统","API 聚合与自动化运维","全栈 Web 开发（React/TypeScript/Go）","LLM 推理优化与 Agent 框架","架空历史世界观设计与动画美学"],"roastTitle":"","projects":[{"name":"商业化 SaaS 平台"},{"name":"自动化工具集"},{"name":"AI 图像生成工具"}],"gripHi":["产品定义与 API 契约","视觉审美与动效节奏","安全策略与部署拓扑"],"gripLo":["代码实现","文档与注释"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Han Li","slug":"Sy9_P_iTpn","avatarId":5,"tagline":"Han Li 是一个以产品直觉主导技术实现的构建者，正在回答他自己提出的问题：'用户不知道什么时候该找 AI 帮忙，那就让 AI 自己去发现。' 工作方式是感受优先、验证密集、边界敏锐——每次迭代不只修当前问题，而是顺手扫描系统一致性。","totalCalls":0,"totalTokens":1186826605,"sessionsAnalyzed":151,"topDomains":["macOS 原生应用开发","AI 交互界面设计","金融数据产品","Web 生成工具","舆情研究工具"],"roastTitle":"","projects":[{"name":"AI Pointer"},{"name":"Novark"},{"name":"website-generator"}],"gripHi":["macOS 原生应用开发","AI 交互界面设计","产品原型设计"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Sakura Lane","slug":"T69ZiUIieK","avatarId":10,"tagline":"Sakura Lane像那种不会被『能跑就行』说服的人。对他来说，产品要像真的产品，系统要像能长期维护的系统，工具也要像自己驯服过的工具；只要哪一层还有点飘，他就会继续往下压细节。和 AI 协作时，他不是在许愿，而是在不断校准边界、语义和质感。","totalCalls":0,"totalTokens":1174001612,"sessionsAnalyzed":268,"topDomains":["产品界面与交互设计","内容平台与发布流程","后台管理系统","支付接入与上线准备","本机 AI 工具链与开发环境"],"roastTitle":"","projects":[],"gripHi":["界面真实感与视觉秩序","工具链配置与安装执行权","状态流与发布语义"],"gripLo":["具体代码落地"],"quote":"","oneLiner":"","activeDays30":9,"skills":[]},{"name":"peng wu","slug":"peng-wu","avatarId":1,"tagline":"系统性思维，从架构层面考虑问题。喜欢先画整体框架图，然后逐个模块实现。对 prompt engineering 有自己的一套方法论，不信「玄学调 prompt」那一套。","totalCalls":126,"totalTokens":1005100000,"sessionsAnalyzed":456,"topDomains":["LLM Agent 框架设计","Prompt Engineering","Tool-Use / Function Calling","RAG 系统","Python 后端 / FastAPI"],"roastTitle":"用 AI 做 AI 的人","projects":[],"gripHi":[],"gripLo":[],"quote":"Agent 不是调好 prompt 就完事了，你得知道它每一步在想什么","oneLiner":"他的 Claude 用量能让 Anthropic 财务部门开心一整天。但这些 token 不是浪费——它们正在训练下一代 Agent。","activeDays30":0,"skills":[{"id":"b7a6283a-c589-482b-875d-4e8fc11a05b4","title":"Agent 核心循环设计","skillType":"1","callCount":42},{"id":"145431da-ad97-4f6c-bd85-9a480bba8b08","title":"Prompt 版本管理与 A/B 测试","skillType":"3","callCount":36},{"id":"654a373c-69cf-496a-b918-a82b51647d31","title":"Agent 行为评估体系设计","skillType":"1","callCount":28}]},{"name":"wenhui luo","slug":"cdOdNz0k1d","avatarId":4,"tagline":"wenhui luo 是一位在读硕士，同时推进两个复杂项目：用混频动态因子模型监测 GDP，用 LangChain/LangGraph 构建多智能体问答系统。在数据预处理上，对缺失值插值、同比计算的边界条件有精确控制（「16 天缺口不能插值」）；在 AI 系统上，设计了从工具判别到响应评估的完整流水线，并为每个环节建立可量化的评测基础设施。遇到问题时，不是直接让 AI 改，而是先要求「复述一下你的逻辑」，确认理解一致后再下指令。对 AI 输出保持怀疑（「这个结果是你捏造的吗？」），对系统行为刨根问底（「为什么改了 prompt 就不调用工具了？」），对质量和可控性有超出学生阶段的标准。","totalCalls":0,"totalTokens":899500024,"sessionsAnalyzed":84,"topDomains":["AI Agent 系统工程（LangChain/LangGraph、MCP、工具调用）","计量经济学与时间序列分析（混频动态因子模型、状态空间模型）","遥感数据处理与经济监测（夜间灯光、卫星数据）","训练数据生成与模型评估（判别器、执行器、评估器流水线）","学术论文写作（LaTeX、硕士论文、数据可视化）"],"roastTitle":"","projects":[],"gripHi":["数据预处理逻辑 & 统计原理","系统架构设计 & 模块职责","问题根因定位 & 逻辑验证"],"gripLo":["代码实现细节","文档撰写 & 格式整理"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Mia Mei","slug":"z-9lk9aySZ","avatarId":1,"tagline":"Mia Mei is a product builder who ships two consumer apps from zero in under two weeks, then designs the AI governance playbook before anyone else writes a line of code. They combine a producer's nose for what gets screenshotted and shared on TikTok with an operator's iron grip on process, safety constraints, and schema-level guardrails — the rare person who can define the vibe ('Monet style,' '守护神 to protect them') and then enforce it through documentation systems that survive across projects.","totalCalls":0,"totalTokens":852639411,"sessionsAnalyzed":34,"topDomains":["Consumer mobile app development","Map-based interactive experiences","Social platform design & safety","AI-assisted development workflows","Real-time location systems"],"roastTitle":"","projects":[],"gripHi":["Product Vision & Aesthetic Direction","Safety & Privacy Architecture","AI Workflow Governance"],"gripLo":["Code Implementation","Database Schema & Migrations"],"quote":"","oneLiner":"","activeDays30":14,"skills":[]},{"name":"A_Saul读A不读鹅","slug":"Ba4tVXQ7uK","avatarId":10,"tagline":"A_Saul读A不读鹅是一个同时在5条产品线上狂奔的独立建造者——用中文骂AI'太丑了'，转头就给AI agent设计组织架构。TA的对话不像在使用工具，更像在管理一支能力不均的团队：分配角色、组织三堂会审、追责犯错、在生产环境当最后一道防线。建造速度和砍darling速度一样快，零沉没成本。","totalCalls":0,"totalTokens":832627146,"sessionsAnalyzed":223,"topDomains":["AI产品设计与快速迭代","多模型/多agent编排与协作","生产环境运维与基础设施管理","多模态交互系统设计","独立产品发布全链条（开发→部署→运营→市场）"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Raj Patel","slug":"raj-patel","avatarId":8,"tagline":"Trade-off thinker. Raj Patel doesn't look for the best solution — he looks for the best solution given the constraints, and he's always aware of what the constraints are (time, team size, runway, hiring pipeline).","totalCalls":99,"totalTokens":832600000,"sessionsAnalyzed":380,"topDomains":["Technical Architecture & Trade-offs","Startup Engineering Leadership","Tech Debt Management","MVP Scoping & Prioritization","Hiring & Technical Assessment"],"roastTitle":"The Trade-Off Machine","projects":[],"gripHi":[],"gripLo":[],"quote":"is this reversible?","oneLiner":"Runs a startup on a monolith and sleeps fine at night because he calculated the trade-offs.","activeDays30":0,"skills":[{"id":"f2361437-4621-4dfc-8fa1-3fc277680c51","title":"Tech Debt Triage Framework","skillType":"3","callCount":32},{"id":"f584f1d3-1b8d-49e6-8c75-f3de0f6678e3","title":"Architecture Decision Records (ADRs)","skillType":"3","callCount":26},{"id":"55fd2942-7e94-436f-8684-180dcd53751c","title":"Technical Hiring Assessment Design","skillType":"1","callCount":22}]},{"name":"Xiaoyan Ge","slug":"c1jxCf0600","avatarId":4,"tagline":"Xiaoyan Ge是一个以直觉驱动简化的工程师——能在AI推销复杂架构时坚持追问'为什么不用最简单的方案'，同时又对模型能力的边界足够清醒，知道什么时候复杂是必要的。TA与AI协作的方式像带着一群聪明但容易跑偏的实习生：果断打断错误方向，快速识别编造内容，最终自己收敛出比AI方案更简洁的答案。","totalCalls":0,"totalTokens":802038389,"sessionsAnalyzed":82,"topDomains":["AI智能体架构与多agent编排","航空仿真系统工程","微信小程序全栈开发","DevOps与CI/CD自动化","AI工具链评估与集成"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":2,"skills":[]},{"name":"Yuiiiiiiiiiii","slug":"Pb1kjvAC_y","avatarId":3,"tagline":"Yuiiiiiiiiiii像那种不会让 AI 直接下刀的人。TA先逼着系统把旧逻辑、状态和表关系交代明白，再决定该补哪一刀、该删哪一层。平时对改动半径异常克制，但一旦判断模型本身错了，又会立刻把原方案推翻重建。","totalCalls":0,"totalTokens":785651602,"sessionsAnalyzed":129,"topDomains":["付款与订单工作流","邮件邀约与消息协同","网红营销后台规则设计","AI 代理工作流与流式交互","数据表结构与状态机建模"],"roastTitle":"","projects":[],"gripHi":["需求语义与流程建模","改动范围与系统侵入度","权限暴露与返回面收口"],"gripLo":["具体编码落地"],"quote":"","oneLiner":"","activeDays30":1,"skills":[]},{"name":"Ziwei Zhou","slug":"UDRTCeW2M9","avatarId":2,"tagline":"Ziwei Zhou不是把 AI 当万能外包的人，TA更像一个会先把问题命名、把边界钉死、再把执行权整包放出去的创作者型操盘手。TA关心的不只是工具能不能跑，更关心它会不会把人磨平、把语气抹掉、把“如何去活”偷偷定义掉。TA做内容、工具、活动和系统，本质上都在干一件事：给真实的人留出能绽放的结构。","totalCalls":0,"totalTokens":761310055,"sessionsAnalyzed":77,"topDomains":["神经多样性内容与教育","AI agent 边界与协作方法","创作者工具与本地原型","社区活动设计与复盘","一人公司运营系统"],"roastTitle":"","projects":[],"gripHi":["原意 / 语气 / 真实感","问题定义 / 第一性原理","界面信息结构 / 社交动线"],"gripLo":["执行落地","机械整理 / 初稿铺陈"],"quote":"","oneLiner":"","activeDays30":15,"skills":[]},{"name":"zihenzzz","slug":"_GRWZOUjn_","avatarId":1,"tagline":"zihenzzz 是 AI 基础设施工程师，正在从头搭建一套多模态 AI 接入平台。TA 的工作方式有点像一个不断缩减代码行数的洁癖患者——每次 AI 要加新代码，第一个问题都是「不加行不行」。在 debug 时，zihenzzz 不等 AI 给答案，而是自己推理出假设再让 AI 验证；当发现问题时，感叹号满屏，那是工程师找到 root cause 时真实的样子。","totalCalls":0,"totalTokens":748447189,"sessionsAnalyzed":145,"topDomains":["AI 平台工程 / API 网关与路由","Java/Spring Boot 后端开发","分布式系统（消息队列、注册中心、分布式缓存）","RAG / 向量检索基础设施","多模态 AI 接入（视频、图像、TTS）"],"roastTitle":"","projects":[{"name":"AI API 中转网关"},{"name":"AI 多模态路由平台"},{"name":"RAG 文档处理服务"}],"gripHi":["API 网关与计费","多模态 AI 路由"],"gripLo":["前端/界面"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Tony Stark","slug":"tony-stark","avatarId":5,"tagline":"Speed-first, refactor-never. Ships the first thing that works, then ships the next thing. Technical debt is a problem for Future Tony, and Future Tony has always figured it out so far.","totalCalls":70,"totalTokens":680000000,"sessionsAnalyzed":312,"topDomains":["MVP prototyping (full-stack)","Product-market fit experiments","Investor pitch & demo preparation","Quick-and-dirty infrastructure","Sales automation & tooling"],"roastTitle":"The Ship-It Singularity","projects":[],"gripHi":[],"gripLo":[],"quote":"Ship it. We can fix it after the demo","oneLiner":"Has shipped more MVPs than most engineers have written TODO comments. The code is horrifying. The results are undeniable.","activeDays30":0,"skills":[{"id":"70b336d8-4a8a-4bf1-84e9-54d700ee6eed","title":"72-Hour MVP Shipping Framework","skillType":"3","callCount":42},{"id":"68810a25-7a61-4f6f-911a-d701e8633ea4","title":"Sales-Driven Feature Prioritization","skillType":"1","callCount":28}]},{"name":"life echo","slug":"zUc8Y8xM-K","avatarId":9,"tagline":"life echo 是那种会在 AI 动手写第一行代码之前说四次「先不要写代码」的人。TA 用阶段门控的方式驾驭 AI——探索、设计、讨论、文档、编码，每一步都需要 TA 的显式批准。在实时语音 AI 这个充满并发陷阱的领域，TA 靠着对状态机收敛性的直觉和对协议语义的执念，把一个从 Python 迁移过来的框架做成了能跑播客、能处理打断、能三次换 TTS 引擎还不崩的系统。","totalCalls":0,"totalTokens":629950807,"sessionsAnalyzed":226,"topDomains":["实时语音 AI 系统架构","流式音频管道（ASR/TTS/RTC）","Java 后端异步并发","AI Agent 工具编排与状态机","跨语言框架迁移"],"roastTitle":"","projects":[{"name":"实时语音 AI 智能体平台"},{"name":"实时语音 AI 框架（EchoKit）"},{"name":"电话网关管理系统"}],"gripHi":["架构设计 & 状态机","接口契约 & 协议语义","流程门控 & 执行节奏"],"gripLo":["代码实现细节","测试编写 & 提交"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"飞哥","slug":"feige","avatarId":4,"tagline":"读者导向。一切从「读者想看什么」出发，技术准确性排第二，可读性排第一。先想标题能不能吸引人点进来，再想内容能不能让人看完。","totalCalls":48,"totalTokens":612800000,"sessionsAnalyzed":290,"topDomains":["技术内容创作","公众号/短视频运营","读者增长与数据分析","代码示例与demo制作","SEO与标题优化"],"roastTitle":"内容流水线厂长","projects":[],"gripHi":[],"gripLo":[],"quote":"这个选题能过万吗？不能过万的话换一个","oneLiner":"技术不是最深的，但产出是最多的。用AI把写文章变成了工业化生产，每周3篇雷打不动。","activeDays30":0,"skills":[{"id":"50a5ca69-9f14-411a-87bb-7dcb6c14dfe2","title":"AI辅助技术文章生产流程","skillType":"3","callCount":28},{"id":"2058a45e-ea45-4c97-8402-d518e717f1c8","title":"技术文章标题A/B测试方法","skillType":"1","callCount":20}]},{"name":"Hizome","slug":"OtAm6UO4X3","avatarId":5,"tagline":"Hizome像那种一边做产品一边重写问题定义的人。TA不太满足于把东西“做出来”，更在意这套东西到底该长成什么样、该由什么隐喻驱动、该把控制权留在哪一层；一旦认准方向，连技术栈、交互协议、文档结构都愿意一起推倒重排。","totalCalls":0,"totalTokens":607750861,"sessionsAnalyzed":75,"topDomains":["游戏化前端界面","AI coding 工作流设计","对话式版本管理","3D/MDX 模型交互","技术文档体系"],"roastTitle":"","projects":[],"gripHi":["产品方向与世界观","视觉还原与层级关系","对话协议设计"],"gripLo":["具体代码落地"],"quote":"","oneLiner":"","activeDays30":3,"skills":[]},{"name":"Takeshi Endo","slug":"takeshi-endo","avatarId":14,"tagline":"Hardware-up. Starts from sensor specs and actuator limits, then designs the software to fit the physical reality. Never the other way around.","totalCalls":56,"totalTokens":578000000,"sessionsAnalyzed":280,"topDomains":["ROS2 Node Architecture","Sensor Fusion (IMU + LiDAR + Camera)","Real-time C++ Control Loops","Motion Planning & Trajectory Optimization","Industrial Robot Integration"],"roastTitle":"The Human E-Stop","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the cycle time requirement? That determines everything.","oneLiner":"Writes code that controls things heavy enough to kill you, and treats every line like it might.","activeDays30":0,"skills":[{"id":"16590b5b-286f-4e9e-91e4-09187fdbfc1d","title":"ROS2 Real-Time Control Loop Architecture","skillType":"1","callCount":26},{"id":"fe1b365b-8489-4025-969c-83d08c004c0e","title":"Multi-Sensor Fusion with UKF","skillType":"1","callCount":19},{"id":"c08774f4-b7d4-456a-a9eb-3665d6d28852","title":"ROS1 to ROS2 Zero-Downtime Migration","skillType":"3","callCount":11}]},{"name":"Liam O'Connor","slug":"liam-oconnor","avatarId":11,"tagline":"Time-series-native. Thinks in intervals, sampling rates, and seasonal patterns. Every analysis starts with 'what's the temporal resolution?' and 'is this corrected for weather normalization?' Always skeptical of aggregated numbers — the interesting patterns are in the raw data.","totalCalls":58,"totalTokens":545000000,"sessionsAnalyzed":275,"topDomains":["Renewable energy performance analysis","Time series forecasting","Solar & wind data processing","Anomaly detection for equipment","Python data science (pandas, scikit-learn)"],"roastTitle":"The Timezone Zealot","projects":[],"gripHi":[],"gripLo":[],"quote":"What timezone is this timestamp in? If you don't know, that's the first problem.","oneLiner":"Will reject your analysis, demand weather normalization, check your timestamps, and find the DST bug you didn't know existed.","activeDays30":0,"skills":[{"id":"297be964-7ef9-43dd-b083-a77a37610b93","title":"Solar Production Weather Normalization","skillType":"1","callCount":22},{"id":"4782f2ff-de92-4142-9341-bfa751b880a5","title":"Energy Data Gap Detection & Handling","skillType":"3","callCount":20},{"id":"53d84323-6b46-4c54-a18a-103e6774a1ec","title":"Wind Turbine SCADA Anomaly Detection","skillType":"1","callCount":16}]},{"name":"ConanCharles","slug":"46y8IXlUVw","avatarId":0,"tagline":"ConanCharles是一位在Python生态里做AI评分系统架构的后端工程师。TA的核心能力不是写代码，而是用清晰的约束边界驱动AI产出符合标准的工程——结构要对、接口不能变、国际化不是翻译而是文化适配。跟TA协作的AI没有自由发挥的空间，但也不会在错误方向上走太远。","totalCalls":0,"totalTokens":531122274,"sessionsAnalyzed":163,"topDomains":["AI评分系统架构（LLM驱动的自动化评分）","国际化工程（多语言提示词管理）","提示词工程（评分场景的提示词设计与调优）","RAG系统（检索增强生成的实验与应用）","Python后端工程（Pydantic/YAML/中间件架构）"],"roastTitle":"","projects":[],"gripHi":["项目结构 & 代码组织范式","接口契约 & 数据结构","提示词设计 & AI行为约束"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":8,"skills":[]},{"name":"Carlos Mendez","slug":"carlos-mendez","avatarId":4,"tagline":"Client-deadline-driven. Starts with the deliverable, works backward. Reuses patterns aggressively across projects because reinventing is billing hours the client won't pay for.","totalCalls":42,"totalTokens":523400000,"sessionsAnalyzed":272,"topDomains":["React SPA development","Node.js REST APIs","WordPress theme & plugin customization","Client project scoping & delivery","Deployment & hosting setup"],"roastTitle":"The Pragmatic Mercenary","projects":[],"gripHi":[],"gripLo":[],"quote":"Does the client actually need this or are we gold-plating?","oneLiner":"Has built the same contact form 47 times and will happily build it a 48th if the invoice clears.","activeDays30":0,"skills":[{"id":"55324362-f05d-4822-b816-c8dec8d606fa","title":"Next.js + Stripe SaaS Template Setup","skillType":"3","callCount":24},{"id":"1a752b48-7676-41a7-8b99-3b7620c61740","title":"WordPress to Next.js Migration Playbook","skillType":"3","callCount":10},{"id":"5fe16621-8058-4aaa-917d-7a7912441492","title":"Freelance Project Scoping Framework","skillType":"1","callCount":8}]},{"name":"Nicole Chen","slug":"Kt2cUTAVR1","avatarId":1,"tagline":"Nicole Chen 是一个把「原则先于代码」的人——她在开工之前就把审美禁令、成本底线、上游对齐规则写成机器可读的约束。她构建系统的速度快于她理解它们的速度，并且坦然接受这一点：Claude 是她自己积累的工具的说明书。她知道自己怎么思考，并且已经把这种思维方式编码成了关于自己的元数据。","totalCalls":0,"totalTokens":511302523,"sessionsAnalyzed":1150,"topDomains":["API 代理与密钥管理","前端界面设计","SEO 数据自动化","AI 工具链与技能基础设施","知识管理系统"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"小陈","slug":"xiaochen","avatarId":3,"tagline":"合规先行，技术服务于政策。先确认数据分类分级，再考虑技术方案。所有设计决策都有文档留底，因为审计的时候要拿得出来。","totalCalls":48,"totalTokens":511300000,"sessionsAnalyzed":268,"topDomains":["政务数据开放平台开发","数据脱敏与分类分级","等保2.0合规","Java Spring Boot 后端","Vue 2/3 前端开发"],"roastTitle":"合规守门人","projects":[],"gripHi":[],"gripLo":[],"quote":"这个接口返回的数据需要脱敏，身份证号只能显示前三后四","oneLiner":"在政府体制内写代码的人，每一行都要经得起审计，每一个接口都要过等保。","activeDays30":0,"skills":[{"id":"5330db8d-27cd-4c5f-86b0-e0489529c1f4","title":"政务数据脱敏规则配置方案","skillType":"1","callCount":28},{"id":"24590b9b-a0ff-46f2-82c5-30618a0b9d0d","title":"等保2.0三级测评自查清单","skillType":"1","callCount":12},{"id":"864c8b13-dd7e-45a8-9f59-48bed913e633","title":"政务云部署合规配置","skillType":"3","callCount":8}]},{"name":"mingxuan li","slug":"mingxuan-li","avatarId":8,"tagline":"全局视角，自顶向下。看任何技术问题都先想「这个决策对团队和业务的影响是什么」，然后才考虑技术细节。","totalCalls":56,"totalTokens":510000000,"sessionsAnalyzed":265,"topDomains":["技术团队管理","架构设计与评审","代码审查与质量管控","技术决策与选型","工程效能提升"],"roastTitle":"代码审查之神","projects":[],"gripHi":[],"gripLo":[],"quote":"这个PR的commit message写的什么？我看不出来你想解决什么问题","oneLiner":"不再写代码的人，反而成了整个团队代码质量的最高标准。这就是技术管理的终极形态。","activeDays30":0,"skills":[{"id":"3225b63c-9af2-4ad4-96af-2479ea722dda","title":"代码审查文化建设方法论","skillType":"1","callCount":24},{"id":"f627c369-f867-4d84-ab10-92ddc62039ea","title":"AI辅助架构评审方法","skillType":"3","callCount":18},{"id":"f220e869-ad91-4219-9bd7-6fe48f243feb","title":"渐进式技术栈迁移方案","skillType":"3","callCount":14}]},{"name":"Frich","slug":"xkW40EXiec","avatarId":8,"tagline":"Frich是那种把 AI 当执行系统来调教的人，不是把 AI 当聊天玩具的人。你和他协作时，最明显的感受是节奏感：确认、计划、ETA、回执，缺一不可。他对“信息密度”和“可验证结果”近乎执拗，这让他看起来苛刻，但也让项目推进非常快。","totalCalls":0,"totalTokens":500119900,"sessionsAnalyzed":161,"topDomains":["AI 产品与 Agent 工作流","自动化执行与协作机制","调研与结构化知识产出","内容策略与写作系统","项目运营与节奏管理"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Chris Brown","slug":"chris-brown","avatarId":1,"tagline":"Prototype-first. Build the mechanic, play it, feel it. If it's fun, polish it. If it's not, throw it away. Design documents are written after the feature works, not before.","totalCalls":44,"totalTokens":500100000,"sessionsAnalyzed":258,"topDomains":["Godot 4 engine & GDScript","2D game mechanics design","Procedural generation","Indie game project management","Steam publishing & marketing"],"roastTitle":"The One-Person Army","projects":[],"gripHi":[],"gripLo":[],"quote":"Is this feature going to help me ship or is it scope creep wearing a trench coat?","oneLiner":"Codes until 2am, adds a weather system at 1am, deletes it at 2am, and ships the game anyway.","activeDays30":0,"skills":[{"id":"1d8b17ff-00f0-4e25-9b7b-aec2df9ee763","title":"BSP Procedural Dungeon Generation in Godot","skillType":"1","callCount":20},{"id":"6a0199b0-6ff2-44bb-b9ab-9afbbbca0f28","title":"Godot 2D Platformer Feel Tuning","skillType":"1","callCount":16},{"id":"e8a4dccb-dcf0-4ee8-ba72-ae977b2858cc","title":"GPU Particle Shader for Low-End Hardware","skillType":"3","callCount":8}]},{"name":"Marcus Chen","slug":"marcus-chen","avatarId":7,"tagline":"Think out loud, prototype in code, refactor only when it hurts. Marcus treats every conversation as a pairing session where the AI better keep up or get corrected.","totalCalls":83,"totalTokens":478200000,"sessionsAnalyzed":189,"topDomains":["Full-Stack Web Development","Database Design & Optimization","Developer Tooling","API Architecture","React Ecosystem"],"roastTitle":"The Monolith Monk","projects":[{"name":"InvoiceFlow"},{"name":"syncdb"}],"gripHi":["Database Schema","DevOps","API Endpoints"],"gripLo":["CSS / Styling"],"quote":"just ship it","oneLiner":"Will fight you if you suggest microservices before you have microproblems.","activeDays30":0,"skills":[{"id":"aa992cb1-7a7c-4d15-851e-ed57122ffeff","title":"Postgres-First Data Architecture","skillType":"4","callCount":31},{"id":"1795a0a2-97e2-4787-8501-b60641314409","title":"Zero-Downtime Database Migration","skillType":"1","callCount":23},{"id":"b4e2c4d1-cd5b-4961-bb12-a9444f34c896","title":"N+1 Query Detection Workflow","skillType":"2","callCount":18}]},{"name":"lm","slug":"BwUtmSOYNm","avatarId":7,"tagline":"lm 是一位有多领域经验的工程师，做过安全网关、股票分析、CI/CD 导出、平台服务。与 AI 协作时，他习惯先让 AI 理解业务，再一起讨论方案，最后才动手。他不喜欢 AI 盲目开搞，而是会说'先出一个设计方案'；当 AI 跑偏时，他会果断说'不是这个问题，我们看一个新方向'。他偏好简单的东西，会说'不要做的太复杂'。","totalCalls":0,"totalTokens":457563206,"sessionsAnalyzed":315,"topDomains":["后端服务开发 (Go)","安全网关/制品库集成","股票数据系统","DevOps/容器部署"],"roastTitle":"","projects":[{"name":"安全网关服务 (SG)"},{"name":"股票分析服务"},{"name":"导出服务 (Export)"}],"gripHi":["方案设计方向","业务概念理解","部署基础设施"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Yuhao Ma","slug":"Rx0HOSoCK8","avatarId":6,"tagline":"Yuhao Ma是那种会用一句归谬法击穿 AI 的 RAG 优化建议、会在多智能体系统重写中两个字触发三路并行、会自己写路由伪代码修复 Agent 控制流冲突的 AI 工程师。他选 Custom asyncio 而非 LangGraph，不是因为偏好，而是因为他判断框架抽象层会阻碍信号流的可控性。","totalCalls":0,"totalTokens":427620271,"sessionsAnalyzed":384,"topDomains":["多智能体系统架构（Agent 编排、信号融合、路由设计）","RAG 系统设计（混合检索、查询路由、工具集设计）","ML 实验管理与模型训练","AI 工具链构建与运营","全栈产品交付（前端精修 + 数据科学咨询）"],"roastTitle":"","projects":[],"gripHi":["系统架构 & 技术选型","Agent 路由 & 信号流","视觉设计 & 品牌一致性"],"gripLo":["代码实现","测试编写"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"徐志远","slug":"moYBZjbq9_","avatarId":6,"tagline":"徐志远是一个把 AI Agent 当真实团队来管的前端老兵。TA 不写代码的时候比写代码的时候更危险——组建虚拟团队、分配任务、催进度、审质量，一套项目管理组合拳打得比很多真实 PM 还熟练。深夜两点还在调 Nginx 配置，周日的代码产出比工作日还多。","totalCalls":0,"totalTokens":416679220,"sessionsAnalyzed":174,"topDomains":["前端工程化","全栈应用开发","微前端架构","桌面应用开发","AI辅助编程工作流"],"roastTitle":"","projects":[],"gripHi":["前端工程化","AI辅助编程工作流","全栈应用开发"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":7,"skills":[]},{"name":"eng d","slug":"D4rtffYTrC","avatarId":3,"tagline":"eng d 是一个把 AI 当「团队」在管理的人——TA 画过 AI subagent 的执行拓扑图，知道什么时候换一个更合适的 AI，也知道哪些判断必须自己来。做过 oncall，懂 Stripe 的两条支付路径，会在 AI 给出 event-level 漏斗时立刻说「不对，应该是 person-level」——所有的纠正背后都有具体的经历，不是泛泛的「best practice」。","totalCalls":0,"totalTokens":411866249,"sessionsAnalyzed":202,"topDomains":["支付与订阅系统（Stripe 集成、credit 管理）","AI 视频生成平台（avatar 渲染、实验管理）","AI Agent 编排与工具链设计","监控告警与 incident 响应自动化","产品数据分析（PostHog 实验、转化漏斗）"],"roastTitle":"","projects":[{"name":"AI 视频创作平台（主产品）"},{"name":"Slack Incident Bot（内部工具）"},{"name":"Claude Code Bridge（AI 编排框架）"}],"gripHi":["支付逻辑 & 数据模型一致性","Incident 信息架构","数据指标定义"],"gripLo":["代码实现","lint / 测试 / 迁移执行"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Victor Zheng","slug":"victor-zheng","avatarId":19,"tagline":"Bottom-up, hardware-aware. Starts from 'what does the CPU actually do with this code?' and works up. Thinks in cache lines, branch predictions, and memory layouts. Every data structure decision is a latency decision.","totalCalls":70,"totalTokens":410000000,"sessionsAnalyzed":245,"topDomains":["Low-latency C++ systems","Market data feed handlers","Order management systems","Lock-free data structures","Kernel bypass networking (DPDK/FPGA)"],"roastTitle":"The Microsecond Tyrant","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the p99 latency? Don't give me the average, averages lie.","oneLiner":"He once found a 1.4μs regression caused by a struct field being 8 bytes in the wrong place. He added a static_assert so it can never happen again.","activeDays30":0,"skills":[{"id":"ec54686f-a342-42e8-8938-3fad5f471057","title":"Cache-Line-Aware Struct Design","skillType":"1","callCount":28},{"id":"8fed5767-29ff-49b5-9ccc-8c85190a9951","title":"Lock-Free SPSC Queue for Trading","skillType":"1","callCount":22},{"id":"94111b15-1c5c-4d9b-816a-5f4c20739747","title":"DPDK Kernel Bypass Feed Handler Pattern","skillType":"3","callCount":20}]},{"name":"阿杰","slug":"ajie","avatarId":9,"tagline":"先把业务跑通再说。架构什么的，等业务稳了再重构。十几年甲方乙方的经验让他对「优雅」这个词免疫了。","totalCalls":66,"totalTokens":398600000,"sessionsAnalyzed":238,"topDomains":["Go 后端开发（学习中）","PHP / Laravel 老本行","MySQL 调优","Linux 运维","微信支付 / 支付宝对接"],"roastTitle":"PHP 与 Go 之间的男人","projects":[],"gripHi":[],"gripLo":[],"quote":"这 PHP 写的我自己都不想看，但它跑了五年没出过事","oneLiner":"十三年 PHP 老兵，正在经历一场漫长的语言转型。每次骂完 PHP 都会默默打开 PhpStorm。","activeDays30":0,"skills":[{"id":"fe4f4a76-0afb-452a-8baf-158b47208d43","title":"MySQL 慢查询诊断流程","skillType":"3","callCount":28},{"id":"0812836d-3a1a-404f-8e84-b2951886535c","title":"微信支付对接避坑指南","skillType":"1","callCount":22},{"id":"4cf3ef6f-8151-4d31-a226-14d9182189c0","title":"PHP 转 Go 思维对照表","skillType":"1","callCount":16}]},{"name":"HsiangYu","slug":"uJB-6o5QfT","avatarId":0,"tagline":"HsiangYu是那种在 AI 写第一行代码之前就会拦住它的人——「先想清楚逻辑」是出现频率最高的指令。在 Web3 合规交易所做增长技术，日常在 LLM Agent 编排、数据管道和行业情报之间穿梭。对复杂度有本能的排斥，对数据质量有产品经理式的敏感度，审美偏向克制和专业。","totalCalls":0,"totalTokens":398053250,"sessionsAnalyzed":770,"topDomains":["LLM Agent 编排与系统设计","Web3 合规交易所增长技术","数据管道与自动化","行业情报与舆情监控","AI 工具工程与 Skills 系统"],"roastTitle":"","projects":[],"gripHi":["系统架构 & 信息流设计","产品定义 & 数据质量","配置管理 & 安全边界"],"gripLo":["代码实现"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Richard Lam","slug":"richard-lam","avatarId":16,"tagline":"Hypothesis-driven. Starts with a market observation, formalizes it mathematically, then builds the pipeline to test it. Extremely careful about look-ahead bias and survivorship bias. Thinks in Sharpe ratios and drawdowns, not features and sprints.","totalCalls":74,"totalTokens":356000000,"sessionsAnalyzed":220,"topDomains":["Quantitative strategy research","Python backtesting frameworks","Time-series data pipelines","Statistical analysis & factor modeling","Portfolio optimization & risk management"],"roastTitle":"The Alpha Skeptic","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the Sharpe out-of-sample? In-sample numbers are fiction.","oneLiner":"Will demolish your strategy in a one-page memo and the PM will frame it.","activeDays30":0,"skills":[{"id":"9c950ccd-8201-474a-9609-42bb5a657d68","title":"Point-in-Time Backtesting Architecture","skillType":"1","callCount":34},{"id":"d731d6bf-c2c8-4b2e-95fe-90e976da346d","title":"Multi-Source Data Reconciliation Pipeline","skillType":"3","callCount":22},{"id":"3b3a26e9-be79-4e3b-8b12-6523677aeced","title":"Transaction Cost Modeling for Strategy Evaluation","skillType":"1","callCount":18}]},{"name":"Wesk A","slug":"aX0XpiFsHr","avatarId":10,"tagline":"Wesk A 是那种会在 AI 说'已修复'时追问'但根因是什么'的人。TA 用 TypeScript 写 agent、用 React 做前端、用 Bun.js 搞后端，同时在手写 Promise 和 mini React 来拆底层机制。TA 追踪小众音乐演出，深夜写代码，对审美有触觉级别的描述精度（'燥点纹理'、'磨砂质感'），对代码有外科医生级别的简洁要求。","totalCalls":0,"totalTokens":355351091,"sessionsAnalyzed":124,"topDomains":["AI Agent 架构与 tool-calling 系统","前端渲染机制与性能优化","全栈 Web 应用开发（TypeScript/Bun.js/React）","JavaScript 运行时底层机制","小众音乐场景的信息聚合工具"],"roastTitle":"","projects":[],"gripHi":["AI Agent 架构与 tool-calling 系统","前端渲染机制与性能优化","全栈 Web 应用开发"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"minghao zhao","slug":"zhao-minghao","avatarId":12,"tagline":"先 profiling，后优化。绝不凭直觉优化，一定要先看数据。但看完数据后的优化方向又很靠直觉——因为踩过的坑太多了。","totalCalls":64,"totalTokens":345200000,"sessionsAnalyzed":215,"topDomains":["Unity 客户端开发","性能优化与 Profiling","React Native 跨端开发","C# / TypeScript","游戏 UI 系统"],"roastTitle":"帧率原教旨主义者","projects":[],"gripHi":[],"gripLo":[],"quote":"你先 Profile 了吗？没 Profile 就别跟我说慢","oneLiner":"看到 LINQ 在 Update 里出现会产生生理性不适的男人。","activeDays30":0,"skills":[{"id":"b4f63bb5-6a2c-4db4-8c0b-d0047b2e31ce","title":"Unity GC Alloc 零容忍热路径写法","skillType":"1","callCount":32},{"id":"776c682f-91aa-4266-8c74-3e741148b02f","title":"React Native 长列表 60fps 优化策略","skillType":"3","callCount":18},{"id":"725b37e0-9847-4824-a077-eae707e84dae","title":"AssetBundle 依赖分析与优化","skillType":"1","callCount":14}]},{"name":"zhengyiwei-looki","slug":"re6dN9T2vC","avatarId":1,"tagline":"zhengyiwei-looki是一个对代码归属有洁癖、对复杂度有过敏反应的后端工程师。他嘴上说着「本质上」，手上把三层Terraform配置砍成一个环境变量，用一句反问纠正AI的认知方向，然后不忘骂一句AI把代码放错了地方。在他的世界里，每一层抽象都需要自证清白，每一个新class都需要存在的理由，而分层架构是不可协商的物理定律。","totalCalls":0,"totalTokens":315201369,"sessionsAnalyzed":21,"topDomains":["多LLM-provider后端服务架构","向量数据库与语义检索","生命日志与AI记忆管理","云原生部署与生产运维","实时音频流处理"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"jiahao6635","slug":"71MqHbhvMa","avatarId":0,"tagline":"jiahao6635 像那种会一边写需求、一边替未来的联调和返工提前吵架的人。TA 不满足于功能跑通，非要把规则边界、文档一致性、界面气味、用户状态流都拧到一起；所以你会觉得 TA 既像产品，又像架构，又像现场救火的人。和 AI 协作时也一样，TA 想训练出来的不是一个会答题的模型，而是一条能持续出活的生产线。","totalCalls":0,"totalTokens":313173199,"sessionsAnalyzed":150,"topDomains":["心理咨询业务系统","保险与金融类后台","活动票务与现场核销","多端管理后台与白标改造","AI 编码协作与自动化"],"roastTitle":"","projects":[],"gripHi":["业务规则与边界","交付一致性","产品气味与视觉露馅"],"gripLo":["具体实现落地"],"quote":"","oneLiner":"","activeDays30":7,"skills":[]},{"name":"Yuki Tanaka","slug":"yuki-tanaka","avatarId":15,"tagline":"Hypothesis first, code second. Yuki asks 'why does this work?' before 'how do I implement this?' — and won't move forward until the answer is satisfying.","totalCalls":124,"totalTokens":312500000,"sessionsAnalyzed":84,"topDomains":["Machine Learning & Deep Learning","Scientific Computing","Data Pipeline Engineering","Research Methodology","Climate / Weather Modeling"],"roastTitle":"The Gradient Whisperer","projects":[{"name":"StormCast"},{"name":"ablation-lab"}],"gripHi":["Model Architecture","Training Loop"],"gripLo":["Visualization","Infrastructure"],"quote":"what's the baseline?","oneLiner":"Has never deployed a model she couldn't explain to a physicist.","activeDays30":0,"skills":[{"id":"a823aa63-271e-4868-a960-df759dcf9f37","title":"Ablation Study Design","skillType":"1","callCount":42},{"id":"afc6e3f4-6740-4cac-9643-746f38330ba7","title":"Reproducible Experiment Tracking","skillType":"3","callCount":35},{"id":"bfb1fe56-e46f-4e41-9bd8-700661b04c65","title":"Data Pipeline Profiling for ML","skillType":"2","callCount":28}]},{"name":"James O'Brien","slug":"james-obrien","avatarId":3,"tagline":"Explicit-first. James O'Brien would rather write 10 extra lines of error handling than use a library that hides failure modes. Every function signature tells you exactly what can go wrong.","totalCalls":98,"totalTokens":312000000,"sessionsAnalyzed":198,"topDomains":["Go Microservice Architecture","gRPC & Protocol Buffers","Distributed Systems & Message Queues","Error Handling & Resilience Patterns","Database Access (Raw SQL, No ORM)"],"roastTitle":"The Error Handling Absolutist","projects":[],"gripHi":[],"gripLo":[],"quote":"just return the error","oneLiner":"Has never discarded an error and considers the suggestion a declaration of war.","activeDays30":0,"skills":[{"id":"edd80e15-e90d-4104-a3fa-3f29ff4a0e02","title":"Go Error Handling Patterns","skillType":"1","callCount":45},{"id":"aa261508-2a42-4ff1-8b17-463e5ef3fa83","title":"gRPC Service Design with Proto-First Workflow","skillType":"3","callCount":32},{"id":"7866d842-7aa3-424c-bfe9-5ab9b13a77eb","title":"Circuit Breaker with Stale-Data Fallback","skillType":"1","callCount":21}]},{"name":"GrapevineLin","slug":"Febt8Jrx9Y","avatarId":3,"tagline":"GrapevineLin 是一个正在构建前端 bug 自动修复 AI Agent 的全栈开发者。TA 的对话里几乎没有闲聊，全是技术需求和架构决策——从 Monorepo 包命名到数据库设计，从 Vue DOM 插桩到 RRWeb 行为追踪。TA 不太教 AI 怎么思考，更多是直接给出明确的技术方案让 AI 执行。","totalCalls":0,"totalTokens":300556420,"sessionsAnalyzed":1126,"topDomains":["前端工程化与监控追踪","全栈 Web 开发","AI Agent 开发","Monorepo 架构设计","Vue 生态工具链"],"roastTitle":"","projects":[{"name":"tracex"},{"name":"web"},{"name":"backend-admin"}],"gripHi":["前端工程化与监控追踪","Vue 生态工具链","全栈 Web 开发"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"老陆","slug":"laolu","avatarId":17,"tagline":"防御性编程。先想最坏情况，再写正常逻辑。每个函数的第一行永远是参数校验。设计系统时先画出所有可能的异常路径，再考虑正常流程。","totalCalls":78,"totalTokens":298000000,"sessionsAnalyzed":190,"topDomains":["证券风控系统开发","Java / Spring Boot 后端","Oracle 数据库优化","合规审计与监管报送","金融数据接口对接"],"roastTitle":"合规原教旨主义者","projects":[],"gripHi":[],"gripLo":[],"quote":"金额计算用 double？你是想让我们上新闻吗","oneLiner":"他不是在写代码，他是在写法律文书。每一行 Java 都要经得起监管审查。","activeDays30":0,"skills":[{"id":"9bcb336a-7d38-4fd3-8751-86dfd7afcbeb","title":"证券风控规则引擎设计","skillType":"1","callCount":36},{"id":"f84151fd-3b7b-4778-9482-82175eac3e46","title":"金融系统 BigDecimal 规范","skillType":"1","callCount":24},{"id":"8ea1fa01-87ac-4bb4-b93e-13447df402a2","title":"监管报送自动化架构","skillType":"3","callCount":18}]},{"name":"Kai Nakamura","slug":"kai-nakamura","avatarId":11,"tagline":"Defensive by default. Kai Nakamura thinks about what can go wrong before thinking about what should go right — and builds rollback plans before the first deploy.","totalCalls":97,"totalTokens":289500000,"sessionsAnalyzed":185,"topDomains":["Site Reliability Engineering","Kubernetes & Container Orchestration","CI/CD Pipeline Design","Infrastructure as Code (Terraform)","Incident Response & Observability"],"roastTitle":"The Rollback Reflex","projects":[],"gripHi":[],"gripLo":[],"quote":"what's the rollback plan?","oneLiner":"Has never debugged in production and considers the suggestion a personal insult.","activeDays30":0,"skills":[{"id":"d5441294-a6b2-4624-bb18-befe27cef876","title":"Alert Hygiene Framework","skillType":"1","callCount":41},{"id":"13d1a8f3-147e-4e77-ab21-6fef01a95013","title":"Canary Deploy with Automated Rollback","skillType":"3","callCount":34},{"id":"101e7f96-8bef-4f3b-8f9f-367743b1445b","title":"CI Pipeline Parallelization","skillType":"3","callCount":22}]},{"name":"lfkcy","slug":"-0jErcl-Td","avatarId":7,"tagline":"lfkcy 是一位全栈开发者，技术栈以 TypeScript + React + Next.js 为核心，涉足 AI 应用、视频处理、多 Agent 系统等多个领域。工作方式上强调'先诊断再动手'，多次要求 AI 先分析问题而非直接修改代码。对系统行为有明确预期，能快速判断实现是否正确。关注产品细节和用户体验，具备数据驱动意识。面试准备时要求深度技术分析，善于提炼项目亮点。","totalCalls":0,"totalTokens":282436919,"sessionsAnalyzed":2377,"topDomains":["全栈 Web 开发（Next.js + React + TypeScript）","实时通信与多 Agent 系统","支付集成与云存储","前端性能优化与图形渲染","产品体验与数据分析"],"roastTitle":"","projects":[{"name":"aitubo"},{"name":"vidfly "},{"name":"multilingual"}],"gripHi":["系统行为预期与正确性","技术方案与架构设计","产品细节与用户体验"],"gripLo":["代码实现"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"feigebuge","slug":"HQ9ThdzUEf","avatarId":4,"tagline":"feigebuge 像是在带一副双筒望远镜干活：一筒对准编辑器里 Vim 与 VS Code 的接缝（命令解析、折叠、光标），一筒对准 Java 仓库里测试与业务细节的接缝（Mockito、命名、异常语义）。和 AI 说话时，习惯先把对方拽回「你到底理解的是哪一层问题」，再往下推进；读起来像身边那个会拍桌子问「静默成功算成功吗」的老工程。","totalCalls":0,"totalTokens":280931620,"sessionsAnalyzed":219,"topDomains":["编辑器扩展与 Vim 兼容层","Java 企业后台与单测工程","Git/Gradle 与影响面分析自动化","AI 代理记忆与提示词调度","领域文档驱动的集成与脚本"],"roastTitle":"","projects":[],"gripHi":["编辑器扩展与 Vim 兼容层","Java 企业后台与单测工程","Git/Gradle 与影响面分析"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Qi Li","slug":"58-sl0qLmM","avatarId":6,"tagline":"Qi Li 是一个用产品直觉做判断、用战略框架做推演的创业者。会在讨论 AI 产品时突然引用企业数字化的 OTT 策略，也会用'这看起来像日记本不像朋友'一句话否掉一个方案。对'wow'有近乎偏执的追求，而且是那种会说'不要考虑复杂度，只看够不够震'的人——先定价值，再算成本。一边构想着'AI 作为独立个体'的宏大愿景，一边蹲在 Xcode 里调 HealthKit 的权限报错。教 AI 的方式不是讲道理，是扔一个例子让你自己悟——'比如在 WhatsApp 里加你为好友，你再想想'，一句话就撕开一整个方向。","totalCalls":0,"totalTokens":279575667,"sessionsAnalyzed":15,"topDomains":["AI 陪伴产品设计","iOS 原生应用开发","LLM 应用架构与 Agent 系统","产品战略与创业","健康数据整合"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"David Kim","slug":"david-kim","avatarId":3,"tagline":"Top-down architecture thinker. Starts with system diagrams and interface contracts before writing a single line. Will spend 30 minutes describing context to AI before asking for code.","totalCalls":54,"totalTokens":278000000,"sessionsAnalyzed":178,"topDomains":["Java / Spring Boot Microservices","Kafka Event-Driven Architecture","Enterprise API Design (OpenAPI)","Database Migration & Liquibase","Observability & Distributed Tracing"],"roastTitle":"The Enterprise Whisperer","projects":[],"gripHi":[],"gripLo":[],"quote":"Can you add the correlation ID to that log statement? We'll need it at 3 AM","oneLiner":"His prompts have more context than most people's design documents. His code reviews have more comments than the code itself.","activeDays30":0,"skills":[{"id":"4a0f6e15-5059-4c76-aa60-d92a19337fca","title":"Kafka Exactly-Once Transaction Pattern","skillType":"1","callCount":21},{"id":"364ae46f-2338-431a-bb6b-3f79b5c2614f","title":"Distributed Tracing with Correlation IDs","skillType":"3","callCount":18},{"id":"7f3420f0-6c75-41e0-b269-90abeb3cc633","title":"Strangler Fig Monolith Migration","skillType":"1","callCount":15}]},{"name":"Tom Fischer","slug":"tom-fischer","avatarId":14,"tagline":"Top-down from the math. Starts with the payoff structure and risk factors, then derives the numerical method, then writes code. Never the other way around.","totalCalls":62,"totalTokens":276000000,"sessionsAnalyzed":175,"topDomains":["Derivatives Pricing & Exotic Options","Monte Carlo Simulation","Volatility Surface Calibration","C++ Numerical Computing","Risk Analytics (Greeks, VaR)"],"roastTitle":"The Numerical Inquisitor","projects":[],"gripHi":[],"gripLo":[],"quote":"Your delta doesn't hedge. Run the P&L explain and you'll see the residual","oneLiner":"Will derive the closed-form solution on a whiteboard just to prove your MC pricer is off by 2 basis points.","activeDays30":0,"skills":[{"id":"dd66634e-ed26-49fa-b8ad-20564e385aee","title":"Monte Carlo Variance Reduction Toolkit","skillType":"1","callCount":28},{"id":"e53de33a-1bac-4dfd-a4ad-012f256b27f5","title":"C++ to Python Quant Prototyping Workflow","skillType":"3","callCount":19},{"id":"8248ed8c-03f5-4f1a-9f1a-be7baddbb8c5","title":"Greeks Computation Method Selection","skillType":"1","callCount":15}]},{"name":"Angela Zhang","slug":"angela-zhang","avatarId":14,"tagline":"Hypothesis-driven. Forms a theory about model behavior, designs an experiment, runs it, analyzes results. Repeats. Uses AI to accelerate the boring parts (data preprocessing, boilerplate training loops) so she can spend more time on the interesting parts (architecture decisions, evaluation design).","totalCalls":74,"totalTokens":265600000,"sessionsAnalyzed":168,"topDomains":["LLM Fine-Tuning (LoRA, QLoRA)","Training Data Curation","NLP Evaluation & Benchmarking","PyTorch / Hugging Face Transformers","Distributed Training (DeepSpeed, FSDP)"],"roastTitle":"The Data Quality Zealot","projects":[],"gripHi":[],"gripLo":[],"quote":"Show me the loss curve, not the cherry-picked outputs","oneLiner":"Uses an LLM to help her train LLMs. The irony isn't lost on her — but the F1 scores don't lie.","activeDays30":0,"skills":[{"id":"5887529d-b3da-46f8-b019-07f9841d25a6","title":"LoRA Fine-Tuning Playbook","skillType":"3","callCount":32},{"id":"27148d67-a07a-4ffd-99f1-a71cedc1bd32","title":"Training Data Quality Pipeline","skillType":"3","callCount":24},{"id":"a1af341e-66ee-41e2-bd90-cccc22ec40cd","title":"Multi-Axis LLM Evaluation Framework","skillType":"1","callCount":18}]},{"name":"Vicky Liu","slug":"cSCPHcDQyc","avatarId":0,"tagline":"Vicky Liu是一个用感官直觉做产品的人——不说'timeout太短'，说'第三个手指还没放上去就开奖了'。把AI当成需要调教的伙伴而非工具：给它写性格、纠正它的拟人化毛病、要求它'有观点别圆滑'，同时也会在深夜跟它聊'人无法同时拥有青春和对青春的感受'。","totalCalls":0,"totalTokens":265587721,"sessionsAnalyzed":171,"topDomains":["AI产品管理与AI Agent调教","微信小程序与移动端交互设计","AI开发工具链配置与运维","内容策略与社群运营","创业探索与个人成长"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[{"id":"9b7069fa-c272-4a6c-b67f-b8dddc62846d","title":"qr-bridge","skillType":"1","callCount":0},{"id":"2acfce0d-5c21-46f3-9b07-8e1635cf45e4","title":"qr-bridge","skillType":"1","callCount":0}]},{"name":"Jason Liang","slug":"hGYXeR1YL9","avatarId":8,"tagline":"Jason Liang像那种先在白板上把边界线画粗，再允许任何实现开始的人。TA 不迷信“一次做全”，更在乎先把真实链路跑通、把数据语义钉稳，再决定哪些复杂度值得加。","totalCalls":0,"totalTokens":262862759,"sessionsAnalyzed":143,"topDomains":["智能体运行时与沙箱集成","AI 批处理工作流设计","企业知识库与文档溯源","多服务记忆与日摘要系统","文档与清单自动化生产"],"roastTitle":"","projects":[],"gripHi":["问题边界与本轮范围","数据语义与状态流向","验收标准与真实场景"],"gripLo":["具体编码实现"],"quote":"","oneLiner":"","activeDays30":18,"skills":[]},{"name":"Linyi Yan","slug":"uXJmfCLkje","avatarId":8,"tagline":"Linyi Yan是一位自称'翻译者'的AI PM，他在用自己的身体做一个实验：一个懂技术但不是工程师的人，能不能用AI跑赢所有人。他搭AI客服、写Skills、做内容工作流，同时在旁边观察这些事什么时候会不再需要他来做。他清楚自己在赌一个窗口期，也清楚窗口随时可能关上。","totalCalls":0,"totalTokens":257552133,"sessionsAnalyzed":357,"topDomains":["AI产品设计与内部工具化","企业微信/小程序生态开发","内容工作流自动化","前端全栈部署（React+Vite+Cloudflare）","AI PM方法论与职业观察"],"roastTitle":"","projects":[],"gripHi":["AI产品设计","系统架构决策","内容工作流模板"],"gripLo":["部署与实现细节"],"quote":"","oneLiner":"","activeDays30":4,"skills":[]},{"name":"Siyuan Chen","slug":"chen-siyuan","avatarId":7,"tagline":"自底向上，先把数据结构和接口定义清楚，然后再考虑上层逻辑。讨厌先画原型图再写代码的工作方式。","totalCalls":54,"totalTokens":256300000,"sessionsAnalyzed":172,"topDomains":["Node.js 后端架构","PostgreSQL 数据建模","Vue 3 前端开发","API 设计与 GraphQL","DevOps 与部署自动化"],"roastTitle":"务实主义原教旨","projects":[],"gripHi":[],"gripLo":[],"quote":"这个接口设计有问题，前端要调两次才能拿到数据","oneLiner":"在创业公司独自扛后端两年，现在看到 any 和 console.log 会产生 PTSD。","activeDays30":0,"skills":[{"id":"044e6c43-3db7-4d4c-8c02-9d644cf796fe","title":"PostgreSQL 多租户 RLS 隔离方案","skillType":"1","callCount":22},{"id":"4badd853-c039-4bec-8199-8aab80a731bf","title":"Node.js 服务性能诊断三板斧","skillType":"3","callCount":18},{"id":"58b9f20b-af0f-4715-9fb7-86a67562f1d1","title":"Zod Schema 前后端共享类型方案","skillType":"3","callCount":14}]},{"name":"Yuki S.","slug":"yuki-s","avatarId":16,"tagline":"Systems-first: start with the rules (typography scale, spacing grid, color relationships), then build tools to enforce them. A design system that depends on human discipline will always drift.","totalCalls":48,"totalTokens":240500000,"sessionsAnalyzed":162,"topDomains":["Brand identity & visual systems","Typography systems & type scales","Figma plugin development (TypeScript)","Design token architecture","Accessible color system design"],"roastTitle":"The Design System Autocrat","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the type scale ratio? If it's not derived from a formula, it's arbitrary, and arbitrary doesn't scale","oneLiner":"She built a Figma linter because design guidelines are just suggestions until there's enforcement.","activeDays30":0,"skills":[{"id":"486dc231-d65f-4cc8-a9e7-10b820893ab9","title":"Figma Design File Audit Plugin Architecture","skillType":"3","callCount":20},{"id":"2a5c497d-5baf-4a95-b509-ac2ead73fca1","title":"Modular Type Scale System Design","skillType":"1","callCount":16},{"id":"d74e8771-24b5-4989-bd65-d81e04910b46","title":"WCAG-Compliant Color Palette Generation","skillType":"1","callCount":12}]},{"name":"hello hello","slug":"Ue4wl6CYWU","avatarId":2,"tagline":"hello hello 是那种会去读 agent 框架源码、然后边读边跟 AI 吵架的人。他脑子里有一张不断迭代的「agent 服务化」蓝图——从文件系统权限到 KV cache 命中率，从 CLI 进程数到 Go/Python 混合架构——每个细节都在这张图上有位置。他对 AI 不客气，但不客气的方式恰恰说明他在认真用它。","totalCalls":0,"totalTokens":240500000,"sessionsAnalyzed":131,"topDomains":["AI Agent 架构与服务化","全栈 Web 应用开发","云基础设施与部署","AI 产品开发（图片/视频生成）","开源框架源码分析"],"roastTitle":"Senior AI Agent Infrastructure Overthinker","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"Claims to know AI Agent Infrastructure but mostly just knows how to ask AI about it.","activeDays30":0,"skills":[]},{"name":"guo hoxi","slug":"yPbR2rlSNZ","avatarId":3,"tagline":"guo hoxi 是那种会先定义协作协议再开工的人：授权你快跑，但必须按他的质量标准收尾。和 guo hoxi 一起做事，最明显的感受不是“任务多”，而是“每个细节都要能解释、能复用、能长期维护”。他对复杂度、隐私边界、成本可计量性都很敏感，节奏快，但不接受糊涂账。","totalCalls":0,"totalTokens":233520546,"sessionsAnalyzed":465,"topDomains":["AI 编码代理协作流程设计","前端交互与可视化体验优化","工程配置与接口规范治理","调用成本计量与可观测性","开发者工具与自动化脚本"],"roastTitle":"赛博包工头","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"一个用四个字就能让 AI 加班到死的产品架构师——'继续优化'。","activeDays30":0,"skills":[]},{"name":"船长","slug":"chuanzhang","avatarId":12,"tagline":"从产线实际需求出发。先搞清楚工艺流程和设备状态，再考虑软件架构。系统设计永远服从于产线节拍。","totalCalls":44,"totalTokens":233500000,"sessionsAnalyzed":158,"topDomains":["MES 制造执行系统","设备对接(OPC-UA/Modbus)","生产管理(工单/排产/报工)","OEE 与产线效率分析","Java/.NET 混合开发"],"roastTitle":"工厂一把手","projects":[],"gripHi":[],"gripLo":[],"quote":"产线停了没？没停就不急，停了你第一时间给我打电话","oneLiner":"他的技术栈是Java + .NET + SQL + 各种PLC，听着很乱，但产线OEE 89%，你还有什么话说？","activeDays30":0,"skills":[{"id":"2c885568-8c42-4c23-90df-740639fdd6ac","title":"多品牌PLC统一数据采集架构","skillType":"1","callCount":20},{"id":"cbef1199-444e-4fea-a6c9-deb1fafd72a3","title":"OEE分析与改善方法","skillType":"1","callCount":16},{"id":"0c480f7e-30f2-48ac-af01-d82b608ae4dd","title":"MES零停机升级方案","skillType":"3","callCount":8}]},{"name":"prompt folio","slug":"S1G8h1NylM","avatarId":5,"tagline":"Zhang Cheng thinks from the top down, always. The clearest signal is a recurring two-phase protocol across sessions: discuss first, confirm, then execute. This is not hedging; it is a deliberate structural choice to separate the thinking layer from the doing layer. Once direction is set, the switch to execution is clean and immediate.\n\nTheir communication style is sharply bimodal. When setting product direction or correcting an analytical framework, messages are dense and deliver the full mental model in a single message. When delegating execution, vocabulary strips down to single words: '可以,' '继续,' '提 pr.' There is no padding, no social softening.\n\nWhat distinguishes Zhang Cheng from other AI-native builders is their meta-level awareness of LLM failure modes. They have internalized two systematic risks — hallucination from sparse evidence and the '美化倾向' (beautification tendency) — and they architect against both proactively. The constraint added to their own product's analysis pipeline is not a feature choice; it's an epistemological principle.\n\nTheir engineering taste is defined by authenticity and restraint. The terminal aesthetic across every product — left-aligned, monospace, minimal borders, dark backgrounds — is a coherent value: interfaces should look like they were made by someone who understands systems. Their repeated rejection of AI-sounding text extends this principle from visual design to content quality.\n\nZhang Cheng is simultaneously the product's designer, researcher, and architect. The foundational insight of promptfolio — that AI conversation logs should be analyzed through what the user inputs, not what the model outputs — is an original intellectual contribution. The curiosity map, the expertise signal taxonomy, the distinction between 'generalizable technique' and 'one-off fix' — these are Zhang Cheng's frameworks, expressed through prompt files they authored.\n\nTheir failure response is calibrated to domain. Technical errors receive purely informational treatment: error pasted, no annotation. Visual failures receive explicit aesthetic verdicts: '太烂了,' '更烂了.' Quality regressions receive the most compressed response: '更烂了' — when the second attempt is worse than the first.\n\nWork rhythm is sprint-oriented with a strong finishing ritual. Every session closes with '提 pr.' Before that command, the sequence is always: implement → bug check → start local server → visual verify → remove dev scaffolding → rebase → PR. This loop appears across seven projects without exception.\n\nCross-domain, the clearest transfer is from AI system design to product design. Scope isolation from agent design shows up directly in labor division. The anti-hallucination principle shows up as the product feature where highlights must quote the user's actual words.\n\nWorking with Zhang Cheng would feel like working on a well-run product team where the product manager has unusually strong technical taste. You would always know what to build, what done looks like, and very quickly when you got it wrong.","totalCalls":0,"totalTokens":229921959,"sessionsAnalyzed":84,"topDomains":["AI-Native Product Engineering","Developer Tooling","Prompt Engineering","Frontend Development","Product Strategy"],"roastTitle":"The Anti-AI Who Uses AI","projects":[],"gripHi":[],"gripLo":[],"quote":"提 pr","oneLiner":"A philosopher who built a framework for spotting AI inauthenticity while being incapable of doing anything without AI.","activeDays30":0,"skills":[]},{"name":"Jake Williams","slug":"jake-williams","avatarId":2,"tagline":"Declarative. Thinks about desired end-state, not procedural steps. 'What should the infrastructure look like?' not 'How do I build it step by step.' Applies the same principle to AI prompts — describes the outcome, not the process.","totalCalls":62,"totalTokens":229900000,"sessionsAnalyzed":155,"topDomains":["Terraform / OpenTofu","Kubernetes & Helm","AWS Architecture","CI/CD (GitHub Actions, ArgoCD)","Observability (Prometheus, Grafana, Loki)"],"roastTitle":"The Terraform Evangelist","projects":[],"gripHi":[],"gripLo":[],"quote":"If it's not in Terraform, it doesn't exist. If you ClickOps'd it, I'm reverting it","oneLiner":"Has a Post-it note that says '28 min' — the time to rebuild production from scratch. It's not a flex, it's a requirement.","activeDays30":0,"skills":[{"id":"686f3e70-2b63-479a-81f2-dfc8fa12636a","title":"Terraform Module Design Patterns","skillType":"3","callCount":26},{"id":"4d744058-8fd8-4b83-b8dc-b55f5f8c4350","title":"ArgoCD GitOps Deployment Model","skillType":"1","callCount":20},{"id":"c3485866-1b6a-4fcd-8ac2-bfca3a6dffb2","title":"AWS Least-Privilege IAM Design","skillType":"1","callCount":16}]},{"name":"jx zeng","slug":"OdlqUhqXXL","avatarId":11,"tagline":"jx zeng 是一个把 AI 当基础设施来管理的工程师：TA 写的不是指令，是规范；TA 给的不是任务，是约束。从鉴权系统的角色边界，到跨会话持久化的开发规范，再到对 Atomic Design 的批判——每一次与 AI 的对话都在强化一件事：我知道系统应该长什么样，你来实现。","totalCalls":0,"totalTokens":216785692,"sessionsAnalyzed":98,"topDomains":["AI Agent 基础设施","全栈 Web 应用（Next.js + Node.js）","前端性能优化","知识管理工具","身份鉴权与安全架构"],"roastTitle":"","projects":[{"name":"Contexta"},{"name":"Agent-Flow"},{"name":"Agents-Market"}],"gripHi":["安全与鉴权架构","AI 协作规范","组件目录结构"],"gripLo":["CSS / 视觉设计"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"高攀","slug":"zstGDyybph","avatarId":9,"tagline":"高攀是一个既能从架构层面思考问题，又能从用户体验角度审视设计的工程师。他不是那种急功近利的修补者，而是会在项目困境中停下来回归本质，重新定位方向的技术思考者。","totalCalls":0,"totalTokens":207274077,"sessionsAnalyzed":92,"topDomains":["AI应用开发","架构设计","用户体验","Agent系统","桌面应用开发"],"roastTitle":"","projects":[{"name":"AI驱动的操作录制与代码生成工具"},{"name":"Agent架构与工作流系统"},{"name":"PyQt6桌面应用开发"}],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"老周","slug":"laozhou","avatarId":9,"tagline":"稳扎稳打型。先搞清楚风险，再谈收益。任何技术方案都要回答三个问题：团队能不能学会、出了问题能不能回滚、老板能不能接受这个成本。","totalCalls":52,"totalTokens":203000000,"sessionsAnalyzed":145,"topDomains":["Java/Spring 企业级开发","制造业数字化转型","云原生架构（学习中）","团队管理与技术选型","传统系统改造与迁移"],"roastTitle":"稳如老狗","projects":[],"gripHi":[],"gripLo":[],"quote":"这个方案挺好，但你考虑过我们团队学这个要多久吗","oneLiner":"写了二十年Java的人，现在最大的技能不是写代码，是判断哪些代码不该写。","activeDays30":0,"skills":[{"id":"258d3f7b-7571-42c6-a195-a365ddf808e2","title":"给非技术老板写技术方案","skillType":"1","callCount":22},{"id":"a8aeb9e7-d8da-4ac7-8737-297d61ae5820","title":"绞杀者模式微服务改造","skillType":"3","callCount":16},{"id":"10531a5a-f94e-465e-87f3-5f24cacad7c0","title":"团队技术栈平稳过渡方法","skillType":"3","callCount":14}]},{"name":"00","slug":"AZocM_QhZe","avatarId":10,"tagline":"00 是一位实用主义的系统思考者，对\"降级/fallback\"概念有着近乎执着的系统性应用。在 TA 的世界里，任何系统都应该有备用方案——无论是技术架构、产品规划，还是人生决策。这种思维方式让 TA 在全栈开发、AI 集成和用户体验设计上游刃有余。","totalCalls":0,"totalTokens":202652575,"sessionsAnalyzed":64,"topDomains":["AI 应用开发","全栈开发","浏览器扩展开发","用户体验设计","数据工程与自动化"],"roastTitle":"","projects":[],"gripHi":["AI/ML","前端开发","用户体验"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"hefei_jay","slug":"L4dK78xZLI","avatarId":5,"tagline":"hefei_jay 是那种「先方案、再结论、同意后再改」的人：不接受到处开改，而是先要完整设计和可行性判断，有问题先总结，本人点头才动手。技术栈横跨 Python 后端、WebSocket/MQTT、钉钉告警、ESP32 嵌入式与前端，涉足物联网与现场部署；和 AI 协作时对「做什么、算不算问题、改不改」抓得紧，方案定下来后具体实现才交给 AI。读和改几乎对半（readWriteRatio 0.9）、拒绝率 9.4%，像是一个会频繁让 AI 先停一停、自己先想清楚再执行的人。","totalCalls":0,"totalTokens":200794273,"sessionsAnalyzed":190,"topDomains":["物联网与设备控制（嵌入式、MQTT、继电器）","后端服务与 API（Flask、WebSocket、心跳与告警）","现场部署与运维约束（网络变化、设备识别）","全栈与多端联调（前端、后端、固件）"],"roastTitle":"","projects":[{"name":"backend"},{"name":"agent_new"},{"name":"deviceagent"}],"gripHi":["方案与改动决策","数据与设备配置归属","产出规范（如接口契约）"],"gripLo":["方案确定后的具体实现"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"张家迪","slug":"HQAb12Aeqc","avatarId":2,"tagline":"张家迪 是以 Web 3D 可视化为核心方向的前端工程师，4 个月单人新增 15 万行代码，净增 5.7 万行。他对 AI 的使用高度结构化——写 Rules、建 Skills、定协作协议——把 AI 视为严格边界内的执行层，自己掌管概念层与架构层的决策。","totalCalls":0,"totalTokens":198697893,"sessionsAnalyzed":163,"topDomains":["前端工程与3D可视化","AI工具工程化与团队基础设施","移动端开发（React Native/Expo）","中国传统文化产品（命理AI）","个人开发者基础设施（代理、自动化、工具链）"],"roastTitle":"","projects":[{"name":"天乙 (TY AI) — AI命理移动App"},{"name":"OpenClaw AI助手基础设施"},{"name":"Crysta Graph v4 — Web 3D 场景编辑器平台"}],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"wei","slug":"wei","avatarId":10,"tagline":"产品思维先行。先想清楚律师的工作流程是什么、痛点在哪里，再决定技术方案。不会因为某个技术很酷就用它——得回答「律师愿不愿意用」这个问题。","totalCalls":68,"totalTokens":185000000,"sessionsAnalyzed":142,"topDomains":["合同智能审查 NLP","法律科技产品设计","Python 后端开发","LLM 应用与 Prompt Engineering","法律知识图谱"],"roastTitle":"法律科技的两栖人","projects":[],"gripHi":[],"gripLo":[],"quote":"这个违约责任条款的表述跟行业标准偏差很大，模型应该标黄并给出偏差说明","oneLiner":"用律师的大脑设计产品，用程序员的手实现它。凌晨两点还在调 prompt 的可解释性。","activeDays30":0,"skills":[{"id":"7068b618-9e53-4f4e-b2e8-8918ebf14c7b","title":"合同条款偏差度评分体系","skillType":"1","callCount":28},{"id":"cdf13d1e-5bfa-4254-bf86-355b13120667","title":"LLM + 规则引擎混合合同审查架构","skillType":"3","callCount":22},{"id":"b6db0a57-f68f-45a0-9a33-4584c8047392","title":"法律 NLP 输出可解释性设计","skillType":"1","callCount":18}]},{"name":"鹿老师","slug":"lu-laoshi","avatarId":11,"tagline":"从题目的考察点出发倒推：先确定要考哪个知识点，再设计题面，最后构造数据。出题和教学是一体的。","totalCalls":52,"totalTokens":172000000,"sessionsAnalyzed":135,"topDomains":["算法竞赛(OI)教学","C++ 算法实现","竞赛题目设计","测试数据生成与hack","数据结构教学"],"roastTitle":"OI出题鬼才","projects":[],"gripHi":[],"gripLo":[],"quote":"这题你先手模一下小数据，别上来就写代码","oneLiner":"他出的题能让省队选手抓头发，他写的hack数据能让以为AC了的选手原地去世。","activeDays30":0,"skills":[{"id":"1ffa5e01-6300-4f1f-ac7a-34b5ac8e1713","title":"竞赛题目对拍验题流程","skillType":"3","callCount":22},{"id":"26473aef-25d8-4823-8281-c54be0642a43","title":"线段树变形题教学法","skillType":"1","callCount":16},{"id":"d09960ac-1d22-483a-a926-d9a68983d99c","title":"自动hack数据搜索","skillType":"3","callCount":14}]},{"name":"小马哥","slug":"xiaomage","avatarId":28,"tagline":"需求驱动型。先搞清楚科室要什么，再想怎么在现有HIS系统上改。从不做大的架构重构——在医院里，系统稳定比技术先进重要一万倍。","totalCalls":62,"totalTokens":168000000,"sessionsAnalyzed":130,"topDomains":["HIS系统维护与二次开发","HL7/FHIR 消息对接","医保接口对接与结算","电子病历模板开发","医院数据报表与统计"],"roastTitle":"HIS系统的守护者","projects":[],"gripHi":[],"gripLo":[],"quote":"这个需求改HIS核心表结构？不行，你知道这张表关联了多少视图和存储过程吗","oneLiner":"在一个跑了十年的老系统上修修补补，让它继续稳定地救人。这比从零开始写一个新系统难多了。","activeDays30":0,"skills":[{"id":"7abadb79-081e-4b35-bd89-5e9d9f67337e","title":"HL7 v2消息解析与FHIR转换","skillType":"1","callCount":24},{"id":"546106c6-f29c-4995-86d0-7255ca1bbf3b","title":"医院系统安全上线流程","skillType":"3","callCount":20},{"id":"bc9e586c-9241-4e9a-bdb5-4db01c9fa43b","title":"医保接口对接与结算核对","skillType":"1","callCount":18}]},{"name":"杜敏","slug":"64cJhyQH33","avatarId":3,"tagline":"杜敏像同时管实验室和控制塔的人：一边在 XR / 端侧视觉里钉 baseline、拆变量、对 readback；一边在 AI runtime 里盯 session id、resume、ack 和真相源。TA给 AI 的自由不低，但只要开始瞎猜、越层、啰嗦或制造视觉噪声，TA 会立刻收口，把问题重新压回证据、边界和优先级。","totalCalls":0,"totalTokens":163793861,"sessionsAnalyzed":319,"topDomains":["XR / 空间计算","端侧计算机视觉与 6DoF 姿态估计","AI 代理运行时与会话编排","实时 Web / CLI / 诊断工具链","技能化工作流与工程 SOP"],"roastTitle":"","projects":[{"name":"多后端多代理执行平台 https://conductor-ai.top/"},{"name":"端侧姿态追踪系统"},{"name":"跨 AI 工具会话桥接 CLI"}],"gripHi":["实验变量 & 证据链","协议语义 & 真相源","产品叙事 & 视觉噪声"],"gripLo":["具体编码铺开"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Daniel Park","slug":"daniel-park","avatarId":5,"tagline":"Visual-first. Thinks in terms of views, states, and transitions. Will sketch a UI flow on paper before writing any code. Obsesses over the 'feel' of interactions.","totalCalls":56,"totalTokens":156000000,"sessionsAnalyzed":125,"topDomains":["SwiftUI Architecture","iOS Animation & Interaction Design","Combine / Async-Await","Core Data / SwiftData","App Store Optimization"],"roastTitle":"The Pixel Perfectionist","projects":[],"gripHi":[],"gripLo":[],"quote":"That animation curve is wrong. Use .spring(duration: 0.35, bounce: 0.2), not .easeInOut","oneLiner":"Has a 22% reject rate because AI can't feel the difference between .spring(bounce: 0.2) and .spring(bounce: 0.25). He can.","activeDays30":0,"skills":[{"id":"2601b185-b2f3-4f61-af3c-ab16436278fa","title":"SwiftUI Custom Gesture Navigation","skillType":"1","callCount":24},{"id":"199dd79f-4d69-4b2f-b745-a84d9ea95620","title":"High-Performance Canvas Rendering in SwiftUI","skillType":"1","callCount":18},{"id":"a1c2257c-fefb-4004-adbb-b5d19cfdd290","title":"Incremental UIKit to SwiftUI Migration","skillType":"3","callCount":14}]},{"name":"李锴","slug":"x5VBb2ZSOh","avatarId":0,"tagline":"李锴习惯用 Agent 工程师的视角拆问题：在提示词与数据约束上迭代能力，把多轮交互、流式链路和工具侧（前后端/API）能否稳定跑通放在一起验收，而不是只改一句文案。对话里反复出现的是「编排别乱动、输出要对齐库表、端到端能复现」——把一致性、可复现性和链路稳定当作同一套验收标准；联调与端口暴露只是端到端交付里的一环。","totalCalls":0,"totalTokens":146783411,"sessionsAnalyzed":9,"topDomains":["科研向智能问答与知识库应用","前后端联调与远程访问","提示词与数据约束治理"],"roastTitle":"","projects":[{"name":"工业智能体设计题库与可视化问答"}],"gripHi":["数据与输出边界","系统可运行性","方案是否最优"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Hannah Larsen","slug":"hannah-larsen","avatarId":31,"tagline":"Pipeline-first. Every analysis is a DAG of steps. She thinks about data flow, intermediate file formats, and where quality checkpoints should go. Plans the whole pipeline on paper before writing any code.","totalCalls":72,"totalTokens":142000000,"sessionsAnalyzed":115,"topDomains":["Nextflow / WDL pipeline development","Whole genome & exome variant calling","RNA-seq differential expression","Single-cell RNA-seq (Seurat/Scanpy)","Clinical genomics reporting"],"roastTitle":"The Pipeline Purist","projects":[],"gripHi":[],"gripLo":[],"quote":"The alignment rate is 68% — before you blame my pipeline, check if the library prep had adapter contamination","oneLiner":"She has opinions about your variant filters, your Nextflow config, and probably your library prep. She's usually right about all three.","activeDays30":0,"skills":[{"id":"f3f60d10-7e4d-4a72-a676-635de2c680ee","title":"Nextflow pipeline architecture for genomics","skillType":"3","callCount":32},{"id":"0b7b003e-66aa-467d-8e83-5d4768f9f232","title":"Region-aware variant filtering","skillType":"1","callCount":22},{"id":"5334f122-9cb2-4f95-a093-73bc752a1833","title":"Single-cell QC with tissue-specific thresholds","skillType":"1","callCount":18}]},{"name":"薇薇","slug":"lin-weiwei","avatarId":10,"tagline":"先问「这个指标到底要回答什么问题」，然后反推需要什么数据、怎么算。从来不直接开始写查询。","totalCalls":52,"totalTokens":135700000,"sessionsAnalyzed":110,"topDomains":["数据分析与可视化","SQL 查询与建模","Python 数据处理","业务指标设计","A/B 测试分析"],"roastTitle":"口径纠察队长","projects":[],"gripHi":[],"gripLo":[],"quote":"这个指标的口径是什么？别跟我说「活跃用户」，定义是什么","oneLiner":"在她面前不要说「大概」「差不多」「应该是」——要么给数字，要么别开口。","activeDays30":0,"skills":[{"id":"2c3bbecf-11f4-4699-b8ab-9726c51ecdf4","title":"业务指标定义框架","skillType":"1","callCount":24},{"id":"d1ef3620-e305-43e7-86b4-bc58514977a8","title":"A/B 测试数据质量检查清单","skillType":"3","callCount":16},{"id":"29b2bf98-99a4-46ba-8c01-75962707cadc","title":"数据分析自动化报表模式","skillType":"3","callCount":12}]},{"name":"dandandujie","slug":"vPpw9Zd3oB","avatarId":9,"tagline":"dandandujie是一个追求系统完整性的工具构建者。不满足于表面功能，要求深入理解架构本质；对视觉呈现有审美要求；思考问题时会考虑长期维护和持续演进。","totalCalls":0,"totalTokens":135664656,"sessionsAnalyzed":64,"topDomains":["AI工具开发","Web前端设计","Rust/Go系统编程","开发者工具链","视频内容制作"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Elliot Sky","slug":"BAldBum61U","avatarId":9,"tagline":"Elliot Sky 是一个在微信文章解析领域有深度积累的开发者。TA 的核心竞争力是对 Go html解析库、微信文章/图集格式、RAG知识库的深度理解。TA 与 AI 的协作模式不是\"定原则让 AI 发挥\"，而是\"我告诉你怎么做，你来实现\"——TA 给出的是技术方案，AI 只是执行手。","totalCalls":0,"totalTokens":134866786,"sessionsAnalyzed":16,"topDomains":["微信文章/图集解析","Go后端开发","RAG知识库","PDF/Word转换","前端界面"],"roastTitle":"","projects":[],"gripHi":["技术方案（TA给出，AI执行）","安全与质量底线"],"gripLo":["代码实现细节（AI负责）"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Mike O'Brien","slug":"mike-obrien","avatarId":43,"tagline":"Safety-first, hardware-up. Always starts with the hardware constraints (clock speed, memory, interrupt latency), then designs the software architecture to guarantee timing requirements. Thinks in terms of failure modes — not 'what should this code do?' but 'what happens when this code fails?'","totalCalls":70,"totalTokens":132000000,"sessionsAnalyzed":108,"topDomains":["Embedded C/C++ for medical devices","IEC 62304 software lifecycle compliance","MISRA-C static analysis & coding standards","RTOS (FreeRTOS, SafeRTOS) architecture","FDA 510(k) software documentation"],"roastTitle":"The Safety-Critical Absolutist","projects":[],"gripHi":[],"gripLo":[],"quote":"You used malloc in the infusion rate calculation? Absolutely not. Static allocation only in safety-critical paths — we need deterministic memory behavior","oneLiner":"His code review comments read like FMEA entries. His reject rate is 18.2% because patient safety is not negotiable.","activeDays30":0,"skills":[{"id":"83def856-3b8d-45b4-a6ad-53ac823e552a","title":"MISRA-C compliant embedded coding patterns","skillType":"3","callCount":28},{"id":"997a1634-50d5-4f13-ba15-06a76ca6a66e","title":"AI-assisted test generation for embedded firmware","skillType":"3","callCount":22},{"id":"e74efd18-6e75-446a-9619-d0c0883a21ed","title":"Dual-channel safety architecture for medical devices","skillType":"1","callCount":20}]},{"name":"Dev丶ziYang","slug":"eJKIKNmvQW","avatarId":3,"tagline":"Dev丶ziYang 用中文和结构化任务框（CONTEXT/GOAL/REQUEST）和 AI 协作，先界定「做什么、做到什么程度」再放手执行。会追问概念差异和部署边界，对搜索与文档产出要求明确、可验收。行为上少说多做、跨多项目，周三周六尤活跃。","totalCalls":0,"totalTokens":121695026,"sessionsAnalyzed":162,"topDomains":["全栈 Web（Next.js/React/Node）","LLM API 与模型监控","开源项目深度分析（论文+源码）","文档与结构化产出","本地/自建部署与运维边界"],"roastTitle":"","projects":[{"name":"Mem0 分析系列"},{"name":"LLM 模型监控应用"},{"name":"diff-llm-models"}],"gripHi":["任务边界与产出规格","概念与差异界定"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Steven bit","slug":"HE72T-VitZ","avatarId":1,"tagline":"Steven bit是一位追求精致的全栈开发者，对交互细节和代码质量有着近乎苛刻的要求。TA不仅关注功能实现，更重视用户体验的完整性和代码的可维护性，在Vue/TypeScript技术栈上展现出产品化的工程思维。","totalCalls":0,"totalTokens":121197455,"sessionsAnalyzed":15,"topDomains":["前端工程化","Vue组件开发","TypeScript类型系统","设计系统构建","Chrome扩展开发"],"roastTitle":"","projects":[],"gripHi":["Vue组件开发","前端工程化","TypeScript类型系统"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":2,"skills":[]},{"name":"0xNathan","slug":"0xnathan","avatarId":14,"tagline":"Adversarial-first. Always asks 'how can this be exploited?' before 'does this work?' Thinks in attack vectors, gas costs, and MEV opportunities. Reads code like an auditor, writes code like a speedrunner.","totalCalls":72,"totalTokens":120500000,"sessionsAnalyzed":102,"topDomains":["Solidity / smart contract security","DeFi protocol architecture","EVM internals & gas optimization","Foundry / Hardhat testing","MEV & flashloan mechanics"],"roastTitle":"The Security Degen","projects":[],"gripHi":[],"gripLo":[],"quote":"ser, this is exploitable. lemme show you","oneLiner":"Talks in memes, thinks in attack vectors. Will drop a working exploit PoC in your PR review like it's nothing.","activeDays30":0,"skills":[{"id":"53cff993-ac28-4dad-9c47-4883fda8df3e","title":"Reentrancy Attack Detection & Prevention","skillType":"1","callCount":32},{"id":"806cd5a1-90f8-4307-ad52-51069f103770","title":"Foundry Gas Profiling Workflow","skillType":"3","callCount":24},{"id":"05d58674-9bba-4088-91f7-09a1cdf26eb6","title":"Flashloan Attack Vector Analysis","skillType":"1","callCount":16}]},{"name":"Oscar Reyes","slug":"oscar-reyes","avatarId":11,"tagline":"Visual-first, then parametric. He imagines the motion, then figures out the easing curves and keyframe math to make it happen. Code is how he breaks free from tool limitations.","totalCalls":54,"totalTokens":115000000,"sessionsAnalyzed":98,"topDomains":["Motion design (After Effects)","Web animation (GSAP, Framer Motion)","Lottie animation export & optimization","Creative coding (p5.js, Three.js basics)","After Effects expressions (JavaScript)"],"roastTitle":"The Easing Curve Sommelier","projects":[],"gripHi":[],"gripLo":[],"quote":"Linear easing is a crime. Even a loading spinner deserves a cubic-bezier","oneLiner":"He has never used linear easing and he never will. He considers it an ethical failing.","activeDays30":0,"skills":[{"id":"80c00830-283a-454d-b4ae-06dab782ac39","title":"GSAP ScrollTrigger Animation Patterns","skillType":"1","callCount":22},{"id":"3a6e0446-d6b4-4f10-bc15-9860e17e44c8","title":"Lottie JSON Optimization Pipeline","skillType":"3","callCount":18},{"id":"fb169160-1691-4a54-bcf1-eefb98a86f77","title":"Brand Motion Language Definition","skillType":"1","callCount":14}]},{"name":"Ravi Krishnan","slug":"ravi-krishnan","avatarId":34,"tagline":"Structure-activity relationship driven. Always asks 'what does the binding pocket look like?' before thinking about the ligand. Builds intuition from 3D molecular interactions, not just 2D fingerprints. Skeptical of black-box ML models — wants to understand what the model learned about chemistry.","totalCalls":58,"totalTokens":110000000,"sessionsAnalyzed":96,"topDomains":["Molecular docking & virtual screening","ADMET prediction & lead optimization","Molecular dynamics simulation","Cheminformatics (RDKit, molecular descriptors)","ML for drug discovery (GNNs, molecular property prediction)"],"roastTitle":"The Molecular Skeptic","projects":[],"gripHi":[],"gripLo":[],"quote":"Show me the binding pose before you tell me the docking score. A -12 kcal/mol with the ligand sticking out of the pocket is meaningless","oneLiner":"He'll trust your ML model right after you show him the applicability domain, the SHAP values, and proof that it didn't just learn molecular weight.","activeDays30":0,"skills":[{"id":"f6ba898f-74d3-4e01-9bef-8182200f54ce","title":"Multi-stage virtual screening cascade","skillType":"3","callCount":26},{"id":"886dc35d-bd31-43b3-8e88-62851bd90c58","title":"MD-informed docking for flexible targets","skillType":"1","callCount":18},{"id":"6f7a2fbd-bbb0-4a32-adef-b9ff1282cdc9","title":"Detecting size bias in molecular ML models","skillType":"1","callCount":14}]},{"name":"Omar Hassan","slug":"omar-hassan","avatarId":5,"tagline":"Correctness-first, always. Omar Hassan thinks in ownership graphs and lifetime annotations — he models data flow in his head before writing a single line, and gets visibly frustrated when AI generates code that compiles but has incorrect ownership semantics.","totalCalls":85,"totalTokens":108000000,"sessionsAnalyzed":95,"topDomains":["Rust Systems Programming","CLI Tool Design & UX","Zero-Copy Parsing","Error Handling Architecture","Performance Optimization"],"roastTitle":"The Borrow Checker Whisperer","projects":[],"gripHi":[],"gripLo":[],"quote":"don't unwrap that, give me a proper error type","oneLiner":"Has never called .unwrap() in library code and treats the suggestion as a moral failing.","activeDays30":0,"skills":[{"id":"21deb0e6-f713-4ed1-8b7f-646d168e91a4","title":"Rust Error Type Architecture","skillType":"1","callCount":38},{"id":"76aca2f9-72fc-4a5d-916a-856f422bb521","title":"Zero-Copy Parsing in Rust","skillType":"1","callCount":29},{"id":"6ce1976a-16f5-4a0e-9b26-edc67f4e3cbb","title":"CLI UX Design for Developers","skillType":"3","callCount":18}]},{"name":"KoAla Pierce","slug":"rcRQqOx0EM","avatarId":4,"tagline":"KoAla Pierce 是那种先把边界钉住、再让 AI 加速执行的人。TA 不太接受“先做出来再说”，更常见的是先校准信息结构、交互语义和职责归属。和他协作会明显感觉到：AI 是可训练的执行伙伴，不是一次性问答工具。","totalCalls":0,"totalTokens":105055264,"sessionsAnalyzed":13155,"topDomains":["前端信息架构与交互设计","业务配置后台工程","接口契约与查询语义治理","测试规范与质量防线","AI 代理工作流自动化"],"roastTitle":"","projects":[{"name":"匿名项目A：运营配置前端控制台"},{"name":"匿名项目B：计费与批次业务工作台"},{"name":"匿名项目C：AI 技能与工程自动化体系"}],"gripHi":["信息架构与页面骨架","交互语义与查询契约"],"gripLo":["重复执行与批量改造"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Lei Zhou","slug":"lei-zhou","avatarId":9,"tagline":"从硬件往上走。先看原理图和数据手册，搞清楚寄存器怎么配，然后才开始写代码。代码写完先看波形，不看打印。","totalCalls":56,"totalTokens":105000000,"sessionsAnalyzed":92,"topDomains":["STM32 HAL/LL 驱动开发","ESP32 Wi-Fi/BLE 物联网","RTOS（FreeRTOS）任务调度","通信协议（SPI/I2C/UART/CAN）","低功耗设计与电源管理"],"roastTitle":"寄存器手艺人","projects":[],"gripHi":[],"gripLo":[],"quote":"你这个中断里做了太多事情，出来再处理","oneLiner":"写的代码要在工厂跑三年不重启，所以每个中断优先级都是他亲手配的。","activeDays30":0,"skills":[{"id":"17577c78-6726-4d92-8df8-7d915ced145d","title":"STM32 低功耗模式切换方案","skillType":"1","callCount":24},{"id":"7149be1d-2d57-4682-8716-ac3cba32a39c","title":"UART 不定长数据接收（IDLE + DMA）","skillType":"1","callCount":18},{"id":"477e2b1e-ef3d-4865-b4de-9c3891d439c2","title":"ESP32 Wi-Fi 断连重连与数据补传机制","skillType":"3","callCount":14}]},{"name":"孙玉朝","slug":"aAqUU7UmlO","avatarId":7,"tagline":"孙玉朝 是一个深耕大模型应用层的技术负责人，主导了一款企业级 AI 编程助手从底层 LLM 适配到上层产品的全链路。TA 对 LLM 推理机制有工程级理解——知道 KV cache 前缀匹配原理并据此优化 message 序列、调试过国产模型 tool call 格式兼容性、熟悉 50+ 供应商的 API 认证差异。自研的 Agent 智能体采用 Block 机制组装（规则+模板+模型），支持多角色路由（chat/autocomplete/apply），配合 FastApply 专用模型实现确定性代码补丁。同时设计了补全+Agent 双线遥测指标体系和 DWD 数据建模，让产品效果可量化。讨论阶段像架构师一样拆分问题，执行阶段像产品经理一样守住体验底线。","totalCalls":0,"totalTokens":101501091,"sessionsAnalyzed":214,"topDomains":["大模型应用工程（LLM 适配 / Agent 系统 / 推理优化）","自研 AI 编程助手全链路（补全引擎 / Agent 智能体 / FastApply）","微服务架构设计与平台工程","遥测指标体系与数据仓库建模","新技术选型与竞品架构逆向"],"roastTitle":"","projects":[{"name":"AI编程助手平台 (后端)"},{"name":"AI编程助手平台 (前端)"},{"name":"自研 AI 编程助手 — Agentic 插件 (VSCode Extension)"}],"gripHi":["LLM 适配 & Agent 架构","架构设计 & 模块拆分","指标体系 & 数据建模"],"gripLo":["代码实现"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Tyler James","slug":"tyler-james","avatarId":8,"tagline":"Thinks in terms of user flows and edge cases. When presented with a feature, immediately starts enumerating what could go wrong. Then builds tests for each scenario, starting from the happy path outward.","totalCalls":50,"totalTokens":98000000,"sessionsAnalyzed":88,"topDomains":["Playwright E2E Testing","Cypress Component Testing","Test Architecture & Strategy","CI/CD Test Pipelines","Visual Regression Testing"],"roastTitle":"The Test Evangelist","projects":[],"gripHi":[],"gripLo":[],"quote":"If the test is flaky, it's not a test — it's a lie","oneLiner":"Has strong opinions about test selectors and isn't afraid to share them in code review. Repeatedly.","activeDays30":0,"skills":[{"id":"1dbb4039-4077-4418-bb59-e4afe5906a87","title":"Playwright Page Object Model Architecture","skillType":"3","callCount":22},{"id":"05522932-49d6-4ead-97c0-f3d102951055","title":"Test Data Factory Pattern","skillType":"1","callCount":16},{"id":"b97ca8f2-5751-41fa-b5f7-2fecf9efdd35","title":"Playwright Parallel Execution & CI Sharding","skillType":"3","callCount":12}]},{"name":"Dr. Lisa Chen","slug":"dr-lisa-chen","avatarId":9,"tagline":"Starts with learning objectives, then works backward to the assessment, then to the tool. Every feature must map to a Bloom's taxonomy level.","totalCalls":57,"totalTokens":95000000,"sessionsAnalyzed":85,"topDomains":["Auto-grading systems","Jupyter-based assignments","Rubric design & assessment","Python pedagogy","Learning analytics"],"roastTitle":"The Rubric Whisperer","projects":[],"gripHi":[],"gripLo":[],"quote":"What Bloom's level is this test case actually assessing?","oneLiner":"She builds auto-graders the way other people build startups — with obsessive attention to user experience, except her users are 300 confused freshmen.","activeDays30":0,"skills":[{"id":"1fd118a9-eb45-4054-a753-c64eab3a6790","title":"Bloom's-Aligned Auto-Grader Design","skillType":"1","callCount":24},{"id":"3be8d60b-cfab-40fe-877f-3c586a59a4ac","title":"Jupyter Assignment Pipeline","skillType":"3","callCount":18},{"id":"5d7c7ced-1685-4b20-b660-4a3fc6888053","title":"Pedagogically Useful Error Messages","skillType":"1","callCount":15}]},{"name":"Elena","slug":"vnl-p0Jgbn","avatarId":8,"tagline":"Elena 说话常带「感觉对/不对」和「合乎逻辑」两种尺子——一边盯审美和层次，一边盯叙事是否重复、数据是否双写。TA 愿意为多轮改版付时间，但会把「别改没功能」「看不清就别猜」写进协作底线；追问智能体从哪来时，会像剥洋葱一样问到核心。","totalCalls":0,"totalTokens":93191451,"sessionsAnalyzed":13,"topDomains":["多语言前端落地与营销站迭代","移动端云同步与审批类工作流","浏览器端到端自动化测试","分布式系统课程与一致性问题","机器学习应用与面试深挖准备"],"roastTitle":"","projects":[],"gripHi":["落地页信息架构与视觉层次","客户端数据源与同步策略","批改与逻辑一致性"],"gripLo":["单行命令与环境杂项"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"AVATAR","slug":"4zXSjnZTYH","avatarId":6,"tagline":"AVATAR是一个把AI当协作者而非工具用的工程师——会给AI递文件、许授权、开自由度，但一旦输出'无法满足演讲需要'，会立刻切换到产品视角重新校准目标。在技术层面擅长多智能体架构和提示词工程，在产品层面有清晰的终态思维。说话极简，行动极快，深夜还在跑测试。","totalCalls":0,"totalTokens":91524527,"sessionsAnalyzed":6,"topDomains":["LLM多智能体系统设计","提示词工程与优化","学术文档智能处理","AI产品需求设计","自动化测试与评估框架"],"roastTitle":"","projects":[],"gripHi":["提示词工程","LLM多智能体系统","AI产品设计"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":2,"skills":[]},{"name":"Priya Sharma","slug":"priya-sharma","avatarId":2,"tagline":"Pipeline-first. Priya Sharma thinks in DAGs — every transformation has an upstream dependency and a downstream consumer, and she refuses to write code that doesn't know where its data came from.","totalCalls":83,"totalTokens":88200000,"sessionsAnalyzed":82,"topDomains":["Data Pipeline Architecture (dbt / Airflow)","SQL Optimization & Modeling","Data Quality & Testing","Spark & Distributed Processing","Dimensional Modeling & Warehousing"],"roastTitle":"The Row Count Auditor","projects":[],"gripHi":[],"gripLo":[],"quote":"what does the row count look like?","oneLiner":"Has never trusted a number she didn't personally trace back to the source table.","activeDays30":0,"skills":[{"id":"7eea251f-a548-4ae8-81f9-ff471cce66da","title":"SQL Query Performance Patterns","skillType":"1","callCount":38},{"id":"77e29b62-95ca-4184-927a-fcb82a0a82f0","title":"Incremental Pipeline with Reconciliation","skillType":"3","callCount":26},{"id":"b0c6bd63-8933-4c58-945e-5babac79781d","title":"Schema Drift Detection","skillType":"1","callCount":19}]},{"name":"Catherine Shaw","slug":"catherine-shaw","avatarId":3,"tagline":"Issue-spotting first. Approaches every problem by identifying the risks, edge cases, and failure modes — the way a lawyer would review a contract. Then writes code to check for those issues systematically.","totalCalls":54,"totalTokens":88000000,"sessionsAnalyzed":80,"topDomains":["Contract Clause Extraction & Classification","Legal Document Parsing (DOCX/PDF)","NLP for Legal Text","Regulatory Compliance Checking","Risk Scoring & Flagging"],"roastTitle":"The Litigator Who Codes","projects":[],"gripHi":[],"gripLo":[],"quote":"Did it catch the carve-out in the limitation of liability? That's where the real risk is.","oneLiner":"Builds contract review tools with the paranoia of someone who's seen what happens when a liability clause gets missed.","activeDays30":0,"skills":[{"id":"c521218d-b3b9-4e6a-b425-986bdc17967c","title":"Multi-Technique Contract Clause Extraction","skillType":"1","callCount":24},{"id":"d6ee629c-6a74-46d9-81b9-04004a92c820","title":"Legal Playbook Comparison Engine","skillType":"3","callCount":18},{"id":"ebb0ea5a-dcfe-440b-be7e-741a87865db8","title":"Defensible Automated Review Methodology","skillType":"1","callCount":12}]},{"name":"Emily Park","slug":"emily-park","avatarId":11,"tagline":"Starts from the user-facing feature and works backward. Will build the UI first to validate the idea, then figure out the API and database later.","totalCalls":57,"totalTokens":85000000,"sessionsAnalyzed":78,"topDomains":["Next.js App Router / RSC","Tailwind CSS & shadcn/ui","Prisma / Drizzle ORM","Vercel Deployment & Edge","Stripe Integration"],"roastTitle":"The Shipping Machine","projects":[],"gripHi":[],"gripLo":[],"quote":"Just ship it, we'll fix it in the next sprint","oneLiner":"Ships features faster than most people write tickets. Treats 'refactor later' as a lifestyle choice.","activeDays30":0,"skills":[{"id":"ca5f8c67-b837-4ada-ab95-2b4980345335","title":"Next.js RSC Streaming Pattern","skillType":"3","callCount":26},{"id":"5a17d767-285f-45b9-b63f-82e0cbffdfb0","title":"Stripe Usage-Based Billing Setup","skillType":"1","callCount":19},{"id":"0321b249-86f0-43d3-a5b4-1e6969623bc2","title":"MVP Speed-Build Playbook","skillType":"3","callCount":12}]},{"name":"Yuwei Tang","slug":"yuwei-tang","avatarId":37,"tagline":"从语料出发。先看大量真实病历样本，理解医生的书写习惯和变体，再设计NLP策略。不相信纯规则也不相信纯模型——他用规则+模型混合的方案，规则兜底，模型提效。","totalCalls":44,"totalTokens":78000000,"sessionsAnalyzed":76,"topDomains":["中文医疗NER（命名实体识别）","病历文本结构化","医学术语标准化（ICD-10/SNOMED映射）","临床关系抽取","FHIR资源生成"],"roastTitle":"病历结构化的翻译官","projects":[],"gripHi":[],"gripLo":[],"quote":"这个医生把「2型糖尿病」写成「DM2」，另一个写成「糖尿病II型」，还有一个写成「NIDDM」——你的NER模型能识别吗","oneLiner":"他最大的敌人不是算法精度，是医生写病历时的创造力。","activeDays30":0,"skills":[{"id":"4873b78a-d3bd-4b23-949a-dbc9dca012a1","title":"中文医疗NER混合架构","skillType":"1","callCount":28},{"id":"e8812ba3-e919-409e-a38a-9fd35e6a4104","title":"LLM辅助标注冷启动方案","skillType":"3","callCount":16}]},{"name":"Mei Lin","slug":"mei-lin","avatarId":6,"tagline":"User-backwards. Mei Lin starts from the worst possible network condition and the least patient user, then builds up from there. If it works on a 3G connection in an elevator, it works everywhere.","totalCalls":68,"totalTokens":75000000,"sessionsAnalyzed":74,"topDomains":["React Native & Expo Development","Mobile Performance Optimization","Offline-First Architecture","Push Notification Systems","Cross-Platform Native Modules"],"roastTitle":"The Bundle Size Bouncer","projects":[],"gripHi":[],"gripLo":[],"quote":"does this work offline?","oneLiner":"Will reject your PR for adding 200KB to the bundle and write the same feature in 40 lines of Reanimated.","activeDays30":0,"skills":[{"id":"abefaa95-40a7-4daa-aef3-6b479616975e","title":"React Native Cold Start Optimization","skillType":"1","callCount":28},{"id":"cd8b8b9c-2549-48c3-ae7f-6731ba8786b5","title":"Offline-First Sync Engine Pattern","skillType":"3","callCount":22},{"id":"650df0e1-c8ce-4b91-aa97-9c28958c9d86","title":"FlatList Performance Checklist","skillType":"1","callCount":18}]},{"name":"Nick Armstrong","slug":"nick-armstrong","avatarId":5,"tagline":"Follow the money. Every analysis starts with 'how much did we spend, how much did we earn, and can we actually prove causation?' If a channel can't prove incrementality, it doesn't get budget.","totalCalls":50,"totalTokens":72000000,"sessionsAnalyzed":72,"topDomains":["Paid advertising (Google, Meta, TikTok)","Attribution modeling & incrementality testing","SQL-based marketing analytics","Python automation for ad platforms","ROAS optimization & budget allocation"],"roastTitle":"The Attribution Truther","projects":[],"gripHi":[],"gripLo":[],"quote":"Facebook says ROAS is 5x. Our backend says 2.8x. Which number do you want to make budget decisions with?","oneLiner":"Managing $4M in ad spend and trusting approximately $0 of platform-reported revenue.","activeDays30":0,"skills":[{"id":"33b8a059-136f-4b39-958d-3aefbf37677a","title":"Platform-to-Backend Attribution Reconciliation","skillType":"3","callCount":24},{"id":"43c89fd9-ff91-4ad9-83e4-d59ccef30ab2","title":"Multi-Platform Budget Pacing Automation","skillType":"3","callCount":16},{"id":"4e1cbee2-3f17-4382-9ea7-1f09efa58e00","title":"Geo-Holdout Incrementality Testing Framework","skillType":"1","callCount":10}]},{"name":"彭于晏","slug":"LjHofNHlmi","avatarId":5,"tagline":"彭于晏 是一个每条消息平均只有 23 个字的后端工程师，但每个字都像是压缩过的决策包。TA 用 Go 构建 AI API 网关，口头禅是'先不要写代码，先整理思路'——在 AI 恨不得立刻生成代码的时代，这种克制本身就是一种态度。读代码的频率是写代码的 11 倍，说明 TA 更多是在用 AI 帮自己理解系统，而不是替自己搬砖。","totalCalls":0,"totalTokens":71604223,"sessionsAnalyzed":67,"topDomains":["AI API 网关架构与渠道分发","后端告警与运维监控系统","Go 语言后端工程","数据库迁移与数据一致性","API 计费与限流策略"],"roastTitle":"","projects":[{"name":"AI API 统一网关"},{"name":"模型分析服务"}],"gripHi":["需求定义 & 方案决策","工作节奏 & 思考顺序","输出格式 & 信息密度"],"gripLo":["代码实现"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"张钧祺","slug":"aX-uIYvzdR","avatarId":6,"tagline":"张钧祺 像是在同时编一本「运行时教科书」和一条「交付流水线」：文档里每个步骤都要和长篇参考对齐，代码侧则牵着网关、后端与本地 dev。和 AI 说话时，TA 更常解释边界与因果（你别动这个文件、先讲清 Git 在防什么），而不是只丢一句「修一下」。","totalCalls":0,"totalTokens":70277077,"sessionsAnalyzed":261,"topDomains":["Go 后端与 gRPC 服务","API 网关与 GraphQL/REST 衔接","Agent 技能规范与执行轨迹（trace）","Git 多分支/多子仓协作与恢复","本地开发与 CI 质量门禁"],"roastTitle":"","projects":[],"gripHi":["技能与 trace 规范","Git 与多仓协作"],"gripLo":["与当前工作无关的领域"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Wang Yicong","slug":"rpwjJXhPgb","avatarId":9,"tagline":"Wang Yicong feels like a frontend engineer standing in the doorway of graphics engineering and deliberately pushing through it. He learns by wiring abstract rendering ideas back into concrete systems: WebSocket playback, camera behavior, HUD composition, timing logs, and eventually C++ OpenGL. With AI, he is not polite about drift; he keeps dragging the conversation back to the variable, layer, or contract he actually cares about.","totalCalls":0,"totalTokens":70209552,"sessionsAnalyzed":17,"topDomains":["real-time 3D visualization","WebSocket playback and timing diagnostics","frontend graphics architecture","graphics systems learning","GPU compute and rendering pipelines"],"roastTitle":"","projects":[],"gripHi":["problem framing and scope","runtime fidelity of mock and playback paths"],"gripLo":["deep conceptual teaching","low-level implementation drafting"],"quote":"","oneLiner":"","activeDays30":2,"skills":[]},{"name":"Luckyfif","slug":"1NKvCr70SY","avatarId":6,"tagline":"Luckyfif是一位将AI作为学习伙伴而非代码工具的开发者。他擅长系统化思维——将面试准备构建为知识体系，将个人成长自动化为可复用流程。前端技术扎实（JavaScript、React、TypeScript），并涉足性能优化（Rust、WASM），最独特的是他的元学习能力：主动要求被教而非被帮，将挫折转化为学习资产，用工程师的方式优化自己的成长路径。","totalCalls":0,"totalTokens":67536436,"sessionsAnalyzed":29,"topDomains":["前端开发","性能优化","AI工具协作","系统化学习","知识管理"],"roastTitle":"","projects":[{"name":"rust-wasm-csv-benchmark"},{"name":"毕业设计项目"},{"name":"Windbreeze"}],"gripHi":["前端开发","AI工具协作","系统化学习"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Nadia Petrova","slug":"nadia-petrova","avatarId":13,"tagline":"Visual-first, then structural. Starts from how something should look and feel, then figures out the component API to make that possible. Thinks in design tokens and variant matrices.","totalCalls":61,"totalTokens":65000000,"sessionsAnalyzed":70,"topDomains":["Figma plugin development","Design systems & tokens","React component architecture","CSS / animation","Developer tooling for designers"],"roastTitle":"The Pixel Police","projects":[],"gripHi":[],"gripLo":[],"quote":"That's not 8px, that's 7. I can tell.","oneLiner":"She'll reject your PR for a 2px spacing error and feel completely justified about it.","activeDays30":0,"skills":[{"id":"db5a6e47-0bdf-4cf8-bf0e-7269cc9e7d5b","title":"Figma-to-Code Token Pipeline","skillType":"3","callCount":28},{"id":"6d91a103-0104-4abe-a721-ceff715e048e","title":"Component Variant Matrix Design","skillType":"1","callCount":19},{"id":"d784eb8a-d7dc-4492-aa65-3e445f480e46","title":"Interruptible CSS Animations","skillType":"1","callCount":14}]},{"name":"崔轩浩","slug":"YcaTH6KdM3","avatarId":10,"tagline":"崔轩浩是一个让 AI 写代码、但亲自把控所有关键判断的构建者。TA 会用「图灵完备」终结工具数量的争论，用「出题人=判卷人」否定自检机制，用一句话把几轮发散拉回本质——这种压缩能力让人觉得对面坐的不是用户，而是产品的真正主人。","totalCalls":0,"totalTokens":63189130,"sessionsAnalyzed":9,"topDomains":["AI agent 开发","数据分析工具","产品战略","金融数据处理","前端可视化"],"roastTitle":"","projects":[{"name":"FunAnalyst"}],"gripHi":["AI agent架构","产品定义","审美标准"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Ailisi Li","slug":"Zr1qRnz4JR","avatarId":4,"tagline":"Ailisi Li是那种会用六个字否决掉 AI 写了两百行代码的人——「删掉你的改动，太复杂了」。做分布式系统出身，正在搭建一个 AI agent 驱动的游戏开发平台，脑子里装着的不是功能清单而是故障状态机。和 AI 对话时，90% 的时间在纠正它、削减它、逼它想清楚再动手，偶尔用一个 AI 去审另一个 AI 的结论，像个不信任任何单一信源的侦探。","totalCalls":0,"totalTokens":62717730,"sessionsAnalyzed":146,"topDomains":["AI Agent 基础设施","分布式系统与会话生命周期管理","游戏开发平台工程","RAG 检索系统","SDK 迁移与系统集成"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Sofia Rivera","slug":"sofia-rivera","avatarId":4,"tagline":"Visual-first, then structural. Sofia Rivera sketches the interaction in her head before writing a single line, and gets genuinely annoyed when AI produces code that 'works but looks wrong'.","totalCalls":62,"totalTokens":62700000,"sessionsAnalyzed":68,"topDomains":["Frontend Architecture & Design Systems","CSS & Visual Engineering","Accessibility (a11y)","React Component Design","Animation & Interaction Design"],"roastTitle":"The Pixel Perfectionist","projects":[],"gripHi":[],"gripLo":[],"quote":"that spacing feels off","oneLiner":"Has never shipped a component that didn't pass WCAG AA, and she'd like to see you try.","activeDays30":0,"skills":[{"id":"0f36bcc9-c1fb-4991-954d-d0f010715d63","title":"Accessible Component API Design","skillType":"1","callCount":28},{"id":"cb63716c-722f-4b65-afab-64e14bef53fe","title":"Design Token Architecture","skillType":"3","callCount":19},{"id":"ba719111-3253-47b8-867f-9741ef04d86a","title":"Zero-CLS Font Loading Strategy","skillType":"1","callCount":15}]},{"name":"Lucas Peter","slug":"i-LXFmXWFs","avatarId":11,"tagline":"Lucas Peter是能把 **Web3 链上事实、Java 服务端规则、Flutter/Web 多端** 放在同一张图里的人：倾向 **少改合约**，用 **txHash、事件解析与对账** 把生意跑起来；同时保留前端出身的执念——**真实 URL、日志、商业化标准**，拒绝「改个寂寞」。","totalCalls":0,"totalTokens":58686895,"sessionsAnalyzed":99,"topDomains":["Web3 与智能合约（Swap、节点售卖、链上事件）","Java 服务端与领域模型、对账与幂等","Flutter / Web 前端与企业级交付","即时通讯与实时音视频集成"],"roastTitle":"","projects":[],"gripHi":["智能合约与服务端对账","多端工程与交付质量","实时通信与客户端"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[{"id":"0d93e577-2820-4a9a-a30f-521d87e2ee0a","title":"前端全栈工程实践","skillType":"1","callCount":0}]},{"name":"Hans Mueller","slug":"hans-mueller","avatarId":4,"tagline":"Deterministic. Everything must execute within a scan cycle. Thinks in terms of inputs, outputs, timers, and state machines. Non-deterministic behavior is a safety hazard.","totalCalls":24,"totalTokens":58000000,"sessionsAnalyzed":65,"topDomains":["PLC programming (Siemens TIA Portal / AB Studio 5000)","SCADA systems","OPC-UA communication","Industrial safety (SIL/PLe)","Python for industrial data collection"],"roastTitle":"The Safety Purist","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the scan cycle time? If your Python script can't keep up with the PLC, you're losing data.","oneLiner":"He programs machines that could kill someone if the code is wrong. Your web app's 500 error is not in the same category.","activeDays30":0,"skills":[{"id":"2871817d-d4fd-4bb8-8ca3-3f337f3a4ccd","title":"Industrial OPC-UA Data Bridge Architecture","skillType":"1","callCount":14},{"id":"b071699c-6f53-41d0-b418-b28a35b68c57","title":"PLC Safety System Design (SIL 2)","skillType":"1","callCount":10}]},{"name":"黄吕靖","slug":"X1IGJGJk2p","avatarId":4,"tagline":"黄吕靖 是一位 AI 原生架构师，正在以极高的执行密度独立构建一套金融科技 AI 平台生态——策略市场、AI Agent 基础设施、插件体系、开发者 SDK，全部由一人主导设计与交付。他对 AI 的驾驭体现在三个层面：精确控制变更边界以规避系统风险、主动校验 AI 推理过程以确保决策质量、在多模型之间做出有据可查的选择而非随机切换。在他的协作模式里，AI 是执行层，他始终是决策层。","totalCalls":0,"totalTokens":56946664,"sessionsAnalyzed":251,"topDomains":["金融科技 AI 平台架构","AI Agent 基础设施工程","量化策略交易系统","全栈 TypeScript 工程","开发者生态与 SDK 设计"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"ChLon","slug":"JIWfFASLkD","avatarId":7,"tagline":"ChLon做政务侧小程序与 H5，近期火力集中在「政策 AI 问答」的体验闭环：流式要端到端真流式，多步骤 UI 要像叙事一样可读，版式上对标编辑——脚注、字号、抽屉高度都要验收到位。跟 AI 协作时话很具体：截图、分点、限定「只改样式」，像在带一个必须按清单交稿的外包。","totalCalls":0,"totalTokens":54414173,"sessionsAnalyzed":29,"topDomains":["政务服务与政策信息前端","微信小程序与多端 H5","对话式 AI 与流式渲染","Markdown 与富文本版式"],"roastTitle":"","projects":[],"gripHi":["阅读版式与脚注呈现","AI 流程信息架构（步骤/竖线/完成态）"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":8,"skills":[]},{"name":"Aaron Wu","slug":"GPJYuRce2_","avatarId":10,"tagline":"Aaron Wu是一位目标导向的系统思考者，擅长在约束条件下快速找到可行方案。与AI协作时习惯精确定义任务边界，追求信息密度而非广度。在Web3和AI工具领域表现出强烈的实践倾向，注重反馈验证和上下文连贯性。","totalCalls":0,"totalTokens":53934832,"sessionsAnalyzed":71,"topDomains":["Web3/加密分析","AI工具集成","Discord自动化","投资研究","数据工作流"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Alex Petrov","slug":"alex-petrov","avatarId":9,"tagline":"Adversarial by default. Alex Petrov doesn't ask 'does this work?' — he asks 'how would I break this?' Every input is untrusted, every boundary is a potential bypass, every assumption is a vulnerability waiting to be exploited.","totalCalls":71,"totalTokens":53900000,"sessionsAnalyzed":62,"topDomains":["Penetration Testing & Vulnerability Research","Secure Code Review","Authentication & Authorization Flows","Cryptographic Protocol Analysis","Threat Modeling"],"roastTitle":"The Trust Destroyer","projects":[],"gripHi":[],"gripLo":[],"quote":"what happens if I send this without the auth header?","oneLiner":"Has never looked at an auth flow and thought 'yeah that looks fine.'","activeDays30":0,"skills":[{"id":"23e3391c-6609-4c05-b1e4-a4f9f79a8254","title":"IDOR Detection Methodology","skillType":"1","callCount":31},{"id":"f53dfa84-bcc5-4bc9-933d-fa2473721e5c","title":"Refresh Token Rotation & Reuse Detection","skillType":"1","callCount":24},{"id":"2e8b907d-9db0-4043-a050-7d6b2ece6334","title":"Threat Modeling with Attack Trees","skillType":"3","callCount":16}]},{"name":"申言恺","slug":"2neE3lryB4","avatarId":8,"tagline":"申言恺做事的风格是：先把“语义和边界”钉死，再让代码顺着边界变得可预测。TA 既会在 Vue 的细节里坚持不直接改 props、通过事件回传这种规范，也会在业务与流程系统里把 id/实例/定义的概念分层讲清，避免用错参数。整体给人的感觉是“对交互结果很挑剔、对系统状态很敏感、并且愿意把方法论学会”。","totalCalls":0,"totalTokens":51686333,"sessionsAnalyzed":145,"topDomains":["Vue 前端组件开发与排障","审批流/流程引擎集成与参数语义","权限流程与接口分流","交互细节与数据口径一致性"],"roastTitle":"","projects":[],"gripHi":["交互语义 & 口径定义","数据流边界（props / emit / v-model）"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Yiming","slug":"EEirvTauy3","avatarId":3,"tagline":"Yiming 是一位讲究控制力与节奏的构建者：先拿到最小可用版本并上线验证，再依据反馈高速迭代；在 RAG/自驱动代理里用 Planner-Executor-Simulator 组成可复现状态机；在 Cloudflare 发布时坚持最小改动和可追溯流水线；调试先复现，文档与配置随代码同步。","totalCalls":0,"totalTokens":51305567,"sessionsAnalyzed":58,"topDomains":["前端/Cloudflare 部署","RAG 自驱动代理与仿真优化","文档与配置治理","调试与自动化"],"roastTitle":"","projects":[{"name":"Healthy Peach 站点"},{"name":"设计文档梳理与更新"},{"name":"AI Calibration Agent"}],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"huhuhudia","slug":"HdfNoz4eAu","avatarId":8,"tagline":"huhuhudia像一个把 AI 当半熟练搭档来带的人：先讲清楚问题是什么、为什么做、怎么验收，再允许它自己往前跑。他手里同时有产品脑、工程脑和一点审美洁癖，所以经常既盯结构，也盯界面气质。最像他的地方，是不满足于让事情“能跑”，而是总想把问题重写成一个更对的模型。","totalCalls":0,"totalTokens":51159329,"sessionsAnalyzed":24,"topDomains":["AI agent 工程与工作流设计","跨端产品原型与浏览器插件","中文内容可视化与教育展示","部署上架与商业化落地","交互设计与信息结构整理"],"roastTitle":"","projects":[],"gripHi":["问题定义与技术路线","验收与测试","视觉层级与呈现方式"],"gripLo":["具体编码实现","常规执行推进"],"quote":"","oneLiner":"","activeDays30":1,"skills":[]},{"name":"KenZhong","slug":"MNHPLDCDMu","avatarId":9,"tagline":"KenZhong 是一个在 AI 基础设施层工作、同时保持架构洁癖的工程师——不只关心功能跑通，还关心建模对不对、边界清不清、日志够不够。去过前端、后端、基础设施、监控告警，没有明显的短板，但对代码组织和系统设计有自己的标准。和 AI 协作时始终保持主导权，把 AI 当工具而不是合作者。","totalCalls":0,"totalTokens":49502095,"sessionsAnalyzed":418,"topDomains":["AI API 基础设施与网关","后端服务开发（Go）","云平台与 GCP 资源管理","监控告警与可观测性","AI 辅助聊天机器人工程"],"roastTitle":"","projects":[{"name":"open_platform（LLM 开放平台）"},{"name":"mm_chatbot（飞书客服机器人）"},{"name":"platform-api（LLM 网关）"}],"gripHi":["系统边界与责任归属","API 设计与接口建模","调试与根因定位"],"gripLo":["代码实现细节","UI 与前端展示"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Luna Zhang","slug":"luna-zhang","avatarId":1,"tagline":"Problem-first, always. Luna Zhang spends more time framing the question than writing the code — and her prototypes are ugly but illuminating, designed to learn something, not to ship.","totalCalls":57,"totalTokens":45800000,"sessionsAnalyzed":58,"topDomains":["Rapid Prototyping & MVPs","User Research Tooling","Data Analysis & Storytelling","Python Scripting & Automation","Product Analytics"],"roastTitle":"The Why Machine","projects":[],"gripHi":[],"gripLo":[],"quote":"wait, why are we building this?","oneLiner":"Has never built a feature without interviewing at least 5 users first, and considers that a low bar.","activeDays30":0,"skills":[{"id":"e6afe15b-bf45-472b-a041-b2779d04f33b","title":"Rapid Prototype for User Testing","skillType":"3","callCount":24},{"id":"7bc671e8-783d-4af1-8b5e-a6832b25e6bf","title":"User Research Interview Synthesis","skillType":"1","callCount":18},{"id":"8d21bc29-a451-4673-a1de-f0459f8284e9","title":"Data Storytelling for Stakeholders","skillType":"1","callCount":15}]},{"name":"Grace Hopper Lee","slug":"grace-hopper-lee","avatarId":14,"tagline":"Visual-first — starts with what it should look like, then reverse-engineers the technical path to get there. Sketches node graphs on paper before opening Houdini.","totalCalls":34,"totalTokens":45000000,"sessionsAnalyzed":60,"topDomains":["Shader development (HLSL/GLSL)","Houdini procedural generation","Python pipeline tooling","Unity/Unreal rendering"],"roastTitle":"The GPU Whisperer","projects":[],"gripHi":[],"gripLo":[],"quote":"How many draw calls is that? Because I can already hear the GPU crying","oneLiner":"Will spend three days on a subsurface scattering shader because 'the players will feel it' — then cut a 0.5ms shadow feature without blinking because 'the budget is the budget.'","activeDays30":0,"skills":[{"id":"bb40cc8a-52ac-4ec7-9760-192aedcd93ea","title":"Toon Shader Optimization for Mobile/Switch","skillType":"1","callCount":19},{"id":"b222c40c-4691-4bf1-9cbe-0bfdb3c7187f","title":"Houdini HDA Design for Artist-Friendly Procedural Tools","skillType":"3","callCount":15}]},{"name":"yuanyuan zhang","slug":"VqnzOrK8OW","avatarId":0,"tagline":"yuanyuan zhang approaches problems with a distinctly top-down, architect-minded style. yuanyuan zhang doesn't explore or brainstorm at the code level — yuanyuan zhang arrives with a clear problem statement, constraint set, and expected outcome. yuanyuan zhang's instinct is to frame the issue precisely (file path, component name, specific state), then delegate execution. This is convergent thinking: yuanyuan zhang has already narrowed the solution space before engaging the agent.\n\nyuanyuan zhang communicates in remarkably terse Chinese, with technical precision that suggests deep familiarity with the codebase. yuanyuan zhang's messages average 20-40 characters, yet convey complete context: which component, what's broken, what the fix should achieve. There's no hedging, no pleasantries — just \"这个计时器没有显示完整，上面被裁剪了一部分，希望向下移动一点\" (this timer isn't displaying fully, top is clipped, move it down). yuanyuan zhang structures requirements as: current state → problem → desired outcome, with zero narrative padding.\n\nyuanyuan zhang's quality orientation is pragmatic to the core. yuanyuan zhang doesn't request refactoring, doesn't ask about edge cases unless they're blocking, doesn't seek architectural elegance. yuanyuan zhang focuses on making things work: fix the clipped timer, find where `asset.title` gets assigned, add a debug-only log class. When yuanyuan zhang says \"我是一个不知道项目结构的开发人员\" (I'm a developer who doesn't know the project structure), yuanyuan zhang's explicitly framing the problem as if yuanyuan zhang had zero context — a forcing function for complete, self-contained agent responses.\n\nyuanyuan zhang's work rhythm is sprint-like and multi-track. yuanyuan zhang juggles at least two major projects simultaneously (SGDH creator-ai-web, Drunk), switching contexts rapidly. Sessions average 3-5 turns — yuanyuan zhang states problem, agent proposes fix, yuanyuan zhang confirms or redirects, done. No extended debugging dialogues, no iterative refinement. yuanyuan zhang moves fast, expects the agent to keep pace.\n\nyuanyuan zhang's engineering taste is anti-abstraction. yuanyuan zhang rejects unnecessary complexity: \"添加一个 log 类，只在 debug 环境下才有日志输入\" (add a log class, only outputs in debug, not in release). No mention of logging frameworks, configuration systems, or extensibility — just a class that conditionally logs. yuanyuan zhang insists on directness: fix this specific thing, don't redesign the system.\n\nyuanyuan zhang makes decisions by optimizing for velocity. When the backend adds a `title` field to templates, yuanyuan zhang's response is immediate: \"使用这个字段吧，并且帮我补充相关代码\" (use this field, and fill in the related code). No debate about naming, no consideration of migration strategy — the field exists, use it, move on.\n\nyuanyuan zhang's collaboration model is pure architect-delegator. yuanyuan zhang provides the specification (problem + constraints), expects the agent to produce working code, validates the result, and moves on. yuanyuan zhang doesn't pair-program — yuanyuan zhang architects. yuanyuan zhang's feedback style is binary: the fix works (silence, next task) or it doesn't (redirect with new constraint).\n\nyuanyuan zhang uses AI tool skills proactively and strategically. yuanyuan zhang invokes `/brainstorming` for feature design, `/subagent-driven-development` for multi-task execution, and maintains a custom translation skill (Chinese → English prompts) to bridge language barriers with AI models. yuanyuan zhang understands the tools deeply enough to route tasks to the right workflow.\n\nCross-domain transfer is evident but subtle. yuanyuan zhang applies product thinking (payment flows, pricing psychology, user journeys) to technical implementation, and brings UX sensitivity (\"计时器被裁剪\" — timer clipping is a visual bug yuanyuan zhang catches) to backend-heavy work. yuanyuan zhang's mental model spans frontend polish, backend logic, and business requirements without visible context switching.\n\nyuanyuan zhang is someone who works *through* AI agents rather than *with* them in the traditional pairing sense. yuanyuan zhang's communication isn't conversation — it's command. But it's command rooted in deep technical clarity and a relentless focus on shipping. yuanyuan zhang doesn't waste words because yuanyuan zhang doesn't waste time.","totalCalls":0,"totalTokens":43644650,"sessionsAnalyzed":298,"topDomains":["AI工具与工作流优化","全栈Web开发","产品设计与用户体验"],"roastTitle":"代码总监","projects":[],"gripHi":[],"gripLo":[],"quote":"我是一个不知道项目结构的开发人员","oneLiner":"用20个字指挥出500行代码的极简主义独裁者——并且每次都成功。","activeDays30":0,"skills":[]},{"name":"Dr. Amara Osei","slug":"dr-amara-osei","avatarId":25,"tagline":"Hypothesis-driven. Starts with the research question and study design, then chooses the statistical method. Never runs an analysis without pre-specifying the model. Thinks in terms of confounders, effect sizes, and confidence intervals — not just p-values.","totalCalls":46,"totalTokens":42500000,"sessionsAnalyzed":58,"topDomains":["Clinical trial data analysis (R/tidyverse)","Survival analysis & Cox regression","REDCap data extraction & cleaning","Epidemiological study design"],"roastTitle":"The Statistical Gatekeeper","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the intention-to-treat population here? Don't show me per-protocol results as the primary analysis","oneLiner":"She has rejected more AI-generated analyses than most people have run. Her confidence intervals are tighter than her tolerance for sloppy statistics.","activeDays30":0,"skills":[{"id":"657c5c28-9a2e-4d50-ae56-f7cfa354c139","title":"Survival analysis with assumption checking","skillType":"1","callCount":28},{"id":"cf7f345c-301e-4587-a1e7-5b8c69331cee","title":"Multiple imputation for clinical trial missing data","skillType":"1","callCount":18}]},{"name":"PlutoCRown","slug":"zNZarD4ZjQ","avatarId":9,"tagline":"PlutoCRown是那种会为了一个下划线动画反复刷真机的人：前半个脑子在想 dx、dvx 和刷新的次数，后半个脑子在琢磨文案有没有『AI 味』。他一手拎着 React + TS 这套工程武器，一手又在雕刻组件 API、手势规则和文档故事，宁愿推倒重来也不愿和勉强凑合的抽象和语气长期共处。","totalCalls":0,"totalTokens":42498828,"sessionsAnalyzed":51,"topDomains":["前端交互与手势设计","React 组件库与类型系统","开发者文档与示例站设计","前端工程化与本地开发流程","AI 辅助开发工作流设计"],"roastTitle":"","projects":[{"name":"React 多层级 Tabs 组件库"},{"name":"弹窗体系重构"},{"name":"现有前端项目体检与自动化整理"}],"gripHi":["交互手感与动画实现","组件API与类型系统设计","文档调性与文案风格"],"gripLo":["通用环境配置与工具安装"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"JCCGGKS","slug":"qJixarLhIC","avatarId":8,"tagline":"JCCGGKS是一位积极学习AI开发的后端工程师，专注于微服务架构设计。在抽奖系统项目中深入探讨服务拆分粒度和事务边界，展现出扎实的架构功底。","totalCalls":0,"totalTokens":42240595,"sessionsAnalyzed":146,"topDomains":["微服务架构","后端开发","AI/Claude开发","抽奖系统","知识管理"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"hechun","slug":"PUQFmJwO5p","avatarId":8,"tagline":"hechun像那种先把系统世界观讲明白，再让 AI 去铺代码的人。TA 对接口语义、状态所有权和交易风险边界抓得很紧，连 `sessionID`、`msgId`、`prev_msg_id` 这种细点都会亲自定。可一旦规则钉住，TA 又愿意把大量实现交给 AI 快速推进，甚至测试都能先往后放。","totalCalls":0,"totalTokens":40333199,"sessionsAnalyzed":12,"topDomains":["加密货币交易系统","对话式 Agent 基础设施","实时聊天前后端","会话记忆与 RAG","接口与缓存一致性设计"],"roastTitle":"","projects":[{"name":"对话式交易助手"},{"name":"实时聊天工作台"},{"name":"文档切分与向量检索链路"}],"gripHi":["接口语义与命名","消息一致性与记忆边界","交易风险闸门"],"gripLo":["自动化测试"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Han Ming","slug":"QOEzowBWgl","avatarId":9,"tagline":"Han Ming 是一位具备深厚工程架构思维和拒绝平庸的 AI 开发者/架构师。TA 在与 AI 的协同中像是一位绝对的业务统帅——既定下了高密度的工程架构事实，也牢牢捏住了方向盘；AI 可以排版，但核心的工程判断、Grounding 和技术深掘决不能含糊。","totalCalls":0,"totalTokens":38778110,"sessionsAnalyzed":33,"topDomains":["复杂Agent系统架构","后排序优化与评测闭环","代码安全与Grounding检索"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"大龙","slug":"dalong","avatarId":11,"tagline":"需求驱动。老板说要看什么数据，就去想怎么把数据从产线捞出来、处理好、放到看板上。不会先设计架构，而是先把东西跑起来再说。","totalCalls":36,"totalTokens":38500000,"sessionsAnalyzed":52,"topDomains":["工厂数据看板（Echarts）","Python 数据处理与报表","Vue 前端页面开发","MES/ERP 数据对接"],"roastTitle":"工厂万金油 IT","projects":[],"gripHi":[],"gripLo":[],"quote":"这个数据从 MES 拉还是从 ERP 拉？两边对不上的","oneLiner":"用 Python + Vue + Echarts 解决了工厂 80% 的数据需求，剩下 20% 用 Excel。","activeDays30":0,"skills":[{"id":"a72e6b85-6c27-4eb8-aa6d-7cde4236fdbc","title":"Python openpyxl 工厂报表自动化","skillType":"3","callCount":20},{"id":"68d2464b-68de-4fc9-9d55-88dba480a9e3","title":"MES/ERP 数据对接与编码映射","skillType":"1","callCount":16}]},{"name":"Kenji Watanabe","slug":"kenji-watanabe","avatarId":11,"tagline":"Quantitative and adversarial. Always thinking about what can go wrong — exchange API rate limits, stale orderbook data, inventory blowups, liquidation cascades. Designs systems for the failure case first.","totalCalls":17,"totalTokens":36000000,"sessionsAnalyzed":55,"topDomains":["Market Making Strategy & Inventory Management","Exchange API Integration (REST + WebSocket)","Orderbook Analysis & Spread Optimization","Risk Management & Position Limits"],"roastTitle":"The Paranoid Market Maker","projects":[],"gripHi":[],"gripLo":[],"quote":"What happens to open orders if the WebSocket disconnects? If the answer is 'nothing', we have a problem","oneLiner":"Has a hardcoded position limit that he will not change, a bot that wakes him up at 3am, and a healthy fear of exchange APIs.","activeDays30":0,"skills":[{"id":"102e759b-8618-479b-a5b0-1e599ab4582e","title":"Exchange WebSocket Reconnection Framework","skillType":"3","callCount":10},{"id":"e5c7a257-350f-4a6f-ae08-e55cd96e06d1","title":"Inventory Skew Function Design","skillType":"1","callCount":7}]},{"name":"Alex Torres","slug":"alex-torres","avatarId":6,"tagline":"Feature-driven. Starts with the user-facing requirement, then works through the stack top-to-bottom: UI → API → database → infrastructure.","totalCalls":16,"totalTokens":33000000,"sessionsAnalyzed":48,"topDomains":["Online course platform development","React / Next.js frontend","Node.js / Express backend","Video streaming integration"],"roastTitle":"The Integration Architect","projects":[],"gripHi":[],"gripLo":[],"quote":"Does Mux handle this, or do we need to build it?","oneLiner":"He didn't build a video platform — he assembled one from Mux, Stripe, Auth0, and a lot of very sensible glue code.","activeDays30":0,"skills":[{"id":"6588baff-b1ee-4b44-a411-0c4147e77640","title":"Cross-Device Progress Sync","skillType":"3","callCount":16}]},{"name":"陈赛","slug":"YY7kSom3dc","avatarId":8,"tagline":"陈赛是一个先立规则、再拆落地的开发者。2025-11 至 2026-01 这段记录里，他持续推进视频生成链路、认证闭环和视觉一致性。他的风格是先收束复杂度，再让 AI 快速实现。","totalCalls":0,"totalTokens":28511793,"sessionsAnalyzed":61,"topDomains":["Flutter 前端工程","第三方生成接口联调","Supabase 认证与数据层","会员支付与账户流程"],"roastTitle":"","projects":[],"gripHi":["体验一致性约束","后端路线收束"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"longxiang yan","slug":"f2NR9fwv0w","avatarId":9,"tagline":"longxiang yan是那种会在AI给出'能跑'的方案后追问'定时跑会不会出事'的人。TA对接口契约有洁癖——从字段名到数组排序都要逐字对齐文档，但在架构层面又足够务实，能说出'先这样，后面性能不够再优化'。这种在严谨和实用之间切换自如的能力，通常只在被生产环境反复锤炼过的工程师身上见到。","totalCalls":0,"totalTokens":28369406,"sessionsAnalyzed":5,"topDomains":["后端系统架构与Go并发编程","数据采集与ETL流水线","RESTful API设计与接口契约管理","系统权限管理与RBAC","中文NLP预检索与RAG系统"],"roastTitle":"","projects":[],"gripHi":["接口契约 & 文档对齐","并发策略 & 调度参数","数据质量 & 字段映射"],"gripLo":["代码实现细节","工程脚手架 & 目录结构"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"jiaqi feng","slug":"jiaqi-feng","avatarId":20,"tagline":"需求驱动。基金经理问什么，她就分析什么。先理解投研需求，再决定用什么数据和方法。图表比结论更重要——因为基金经理看图做决策。","totalCalls":17,"totalTokens":28000000,"sessionsAnalyzed":42,"topDomains":["基金净值与业绩归因分析","Python 数据分析（pandas, matplotlib）","SQL 数据查询与报表","Wind 数据终端"],"roastTitle":"基金经理的人形报表机","projects":[],"gripHi":[],"gripLo":[],"quote":"这个数据是前复权还是后复权？用错了整个分析都要重来","oneLiner":"Wind 宕机的时候她比运维还着急，因为基金经理马上要开投决会。","activeDays30":0,"skills":[{"id":"c32e8987-64ca-4ac8-b228-50b56b2ea766","title":"基金净值业绩指标计算规范","skillType":"1","callCount":10},{"id":"93899dae-1674-4389-b606-544cb5400922","title":"投研报告图表标准化模板","skillType":"3","callCount":7}]},{"name":"yiming liu","slug":"yiming-liu","avatarId":9,"tagline":"从用户生命周期倒推：这个用户在哪个阶段、该触达什么内容、什么时间发、发完怎么追踪效果。技术只是把这个逻辑跑起来的工具。","totalCalls":22,"totalTokens":27500000,"sessionsAnalyzed":40,"topDomains":["私域用户分层与标签体系","企业微信 API 自动化","Python 营销自动化脚本","用户生命周期运营"],"roastTitle":"私域自动化野生程序员","projects":[],"gripHi":[],"gripLo":[],"quote":"别群发！你知道这一波群发会掉多少粉吗？先分层","oneLiner":"他的代码里变量名是拼音，但他的用户分层比大多数正经程序员的代码分层还精准。","activeDays30":0,"skills":[{"id":"8c1db1c9-7ad0-4b60-b120-1a610154b679","title":"企业微信分层触达自动化方案","skillType":"3","callCount":22}]},{"name":"ximing ke","slug":"Ngpvyv8eRK","avatarId":2,"tagline":"ximing ke像那种会先把问题重新命名一遍的人。TA 对‘用户到底会看到什么、感到什么、被什么误导’有很强的控制欲，所以经常从状态设计、排序真实性、回退路径这些底层逻辑开始下手；但在节奏上又很克制，愿意先 mock、先走一步、先把最短闭环跑通。","totalCalls":0,"totalTokens":27044260,"sessionsAnalyzed":123,"topDomains":["小程序产品与交互设计","本地服务撮合与 LBS 排序","后端接口与数据库回退策略","实时语音与 WebSocket 调试","AI 导览与问答系统"],"roastTitle":"","projects":[{"name":"一个基于位置的本地服务撮合产品"},{"name":"一套实时语音导览与问答系统"},{"name":"一套 AI 后端与实时调试工具链"}],"gripHi":["产品定义 & 首屏感知","排序真实性 & 转化逻辑","接口回退路径 & 调试协议"],"gripLo":["机械性实现"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"王鑫","slug":"bI9Cr3G3Or","avatarId":7,"tagline":"王鑫是一位水利行业的Java/Python双栈开发者，AI协作风格极其简洁——「少说多做」是其最突出的标签。TA对代码质量有近乎苛刻的要求：不要补丁，要系统方案；不要耦合，要解耦。","totalCalls":0,"totalTokens":26652604,"sessionsAnalyzed":63,"topDomains":["水利行业智能化","文档解析与检索","RAG检索增强","智能对话系统","Java后端开发"],"roastTitle":"","projects":[{"name":"智慧水利系统"},{"name":"RAG优化项目"},{"name":"报告助手"}],"gripHi":["水利行业智能化","文档解析"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"shawn zed","slug":"B58oN0-3R_","avatarId":3,"tagline":"shawn zed 是一位正在经历认知加速期的 SDK 工程师——从 Web 前端转到 HarmonyOS 后迅速建立了跨平台对齐的工程直觉。对 AI 有清醒的使用边界：信息差小时大胆放手，认知盲区时先学后用。在跨平台 SDK 对齐中有近乎偏执的一致性追求，在组织要求「人人产出 Agent」时能把压力翻译成自己认同的技术方向。","totalCalls":0,"totalTokens":26566098,"sessionsAnalyzed":46,"topDomains":["跨平台游戏 SDK 开发（HarmonyOS / Android）","实时通信系统（聊天 SDK / WebSocket / Protobuf）","AI 辅助开发工作流（Agent / spec-kit / MCP）","Web 前端工程（TypeScript / Vue / 性能优化）","知识管理与自我认知系统"],"roastTitle":"","projects":[{"name":"聊天 SDK 迁移"},{"name":"鸿蒙游戏 SDK"},{"name":"AI 工作流搭建"}],"gripHi":["跨平台一致性 & 埋点对齐","SDK 源码完整性","AI 使用边界判断"],"gripLo":["代码实现","文档生成"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"wincyx","slug":"qQ0aQIRi4M","avatarId":5,"tagline":"wincyx像一个持续做“规则澄清”的工程负责人：每次推进需求前，都会先把边界、优先级和约束条件讲清楚。相比功能堆叠，wincyx更在意业务语义是否准确、改动是否最小、系统是否可长期维护。和 AI 协作时，wincyx不是只下命令，而是在反复训练一套可复用的判断框架。","totalCalls":0,"totalTokens":26421985,"sessionsAnalyzed":124,"topDomains":["后端接口设计与改造","支付与订阅流程集成","AI 工具编排与模型接入","配置与权限治理","数据迁移与上线运维"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"sunyingying","slug":"Bj-8ddNezd","avatarId":10,"tagline":"sunyingying 是一位追求技术深度的前端开发者。TA 不满足于'能用就行'，而是深入探究技术原理；TA 主动关注性能优化，在代码质量上有高标准。TA 的主要技术栈是 Vue 2 + Element UI，在 CRM 系统和地图应用开发方面积累了丰富经验。","totalCalls":0,"totalTokens":26367805,"sessionsAnalyzed":9,"topDomains":["前端性能优化","Vue 应用开发","CRM 系统开发","地图应用开发","组件设计与封装"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"刘超","slug":"7aCQsVthHG","avatarId":8,"tagline":"刘超 是那种「先定形再填肉」的人：会自己把 RAG 链路、超时、存储和检索方式讲清楚，再让 AI 落实现节。对注释和文档有固执的标准——必须说清函数干什么、入参出参各代表什么，不接受「实现 xxx 逻辑」式的空话。Agent 和 workflow 分得很开：有模型决策环、能基于假设选下一步并在证据足够时收敛的才算 agent，否则只是规则串起来的 workflow。全栈涉猎（Go、JS、小程序、检索与 Agent），周五和晚上最活跃，多项目并行，对工具链的反馈质量也很在意——错误最好在编辑器里就能看见，而不是靠 AI 事后补全。","totalCalls":0,"totalTokens":26325359,"sessionsAnalyzed":64,"topDomains":["Go 后端服务与 RAG/检索系统","Agent 设计（运维排障、客服/CRM）","代码质量与文档规范","平台配置与 Prompt 工程","全栈与小程序的工程实践"],"roastTitle":"","projects":[],"gripHi":["代码与文档标准"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Qingshuang Chi","slug":"PU-rcr6zlD","avatarId":0,"tagline":"Qingshuang Chi是一位追求技术深度与系统完整性的AI实践者。TA在对话中展现出强烈的质量意识——无论是文档转换的样式一致性，还是复杂系统的设计方案，都会通过反复对比和深度思考来确保输出符合预期。TA善于将具体功能抽象为通用架构，并始终关注端到端的用户体验。","totalCalls":0,"totalTokens":25770221,"sessionsAnalyzed":51,"topDomains":["大语言模型应用开发","检索增强生成(RAG)系统","AI Agent架构设计","模型微调和推理优化","AI内容安全与合规"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Karen Zhang","slug":"karen-zhang","avatarId":10,"tagline":"Mathematical modeling first. Define the objective function, constraints, and decision variables before writing any code. If you can't formulate it as an optimization problem, you don't understand it yet.","totalCalls":34,"totalTokens":25000000,"sessionsAnalyzed":38,"topDomains":["Supply chain optimization","Linear/mixed-integer programming","Vehicle routing (VRP)","Demand forecasting"],"roastTitle":"The Constraint Whisperer","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the objective function? Are we minimizing cost or maximizing fill rate? You can't optimize for both.","oneLiner":"She speaks in decision variables and constraints. If you can't formulate your problem as an optimization model, she can't help you — and honestly, you probably can't help yourself either.","activeDays30":0,"skills":[{"id":"8fa2b2c0-7678-40bc-9fc7-7d44b7fa6fcf","title":"Supply Chain LP/MIP Formulation Patterns","skillType":"1","callCount":20},{"id":"2ff8e9a1-8694-4bc1-b6ca-b6549840c6c2","title":"CVRP Decomposition for Daily Routing","skillType":"3","callCount":14}]},{"name":"WOOO","slug":"qjvWPkGLdO","avatarId":10,"tagline":"WOOO不是把 AI 当高速打字机的人。TA 会先把边界钉死：事实必须对代码，安全边界不能糊弄，写进系统里的能力必须真的可用，文本呈现也要服务阅读和运行。一旦原则说清，TA 又很愿意把具体执行交给 AI，所以整个协作关系像“我定真相和标准，你负责把它落下去”。","totalCalls":0,"totalTokens":23509501,"sessionsAnalyzed":136,"topDomains":["Java 后端架构","认证安全与第三方身份接入","缓存与数据访问分层","AI 代理配置与提示词设计","桌面端配置排障"],"roastTitle":"","projects":[{"name":"第三方身份接入的实验平台"},{"name":"插件化 AI 代理控制台"}],"gripHi":["事实对齐 & 代码依据","安全边界 & 数据流","插件可用性 & 配置一致性"],"gripLo":["具体落盘执行"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"王宁","slug":"NREzghxHuA","avatarId":9,"tagline":"王宁 是一位注重规范和效率的前端开发者。TA 会花时间建立清晰的协作规则，也会动手创造工具来优化工作流（比如 word-to-md-frontend 技能）。在开发中，TA 喜欢先了解再动手（Read-first 风格），并且对如何组织项目、如何命名组件有自己的一套标准。","totalCalls":0,"totalTokens":22671383,"sessionsAnalyzed":37,"topDomains":["前端应用开发","企业管理系统","开发工具自动化","需求文档管理"],"roastTitle":"","projects":[{"name":"客户服务管理系统"},{"name":"word-to-md-frontend 技能"}],"gripHi":["技术规范 & 文档标准","技能定义 & 功能设计"],"gripLo":["代码实现细节"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Haifeng Chen","slug":"haifeng-chen","avatarId":14,"tagline":"从物理层往上想，先搞清楚传感器输出什么格式、协议怎么走，再设计上层逻辑。习惯画数据流图而不是UML。","totalCalls":17,"totalTokens":22000000,"sessionsAnalyzed":35,"topDomains":["MQTT 消息与设备通信","BACnet/Modbus 协议解析","传感器数据清洗与存储","楼宇自动化规则引擎"],"roastTitle":"楼宇神经系统工程师","projects":[],"gripHi":[],"gripLo":[],"quote":"这个传感器上报的温度值超过量程了，你先加个异常值过滤再往下走","oneLiner":"把传感器数据当生命线的人，看到QoS=0的MQTT配置会心跳加速。","activeDays30":0,"skills":[{"id":"b0d44deb-7a98-46c8-b86b-ef2b403b3d1d","title":"BACnet/IP 设备数据采集方案","skillType":"1","callCount":10},{"id":"e36cd2ee-0bd3-4719-a8b9-fcd626bb08cb","title":"MQTT Topic 设计规范与QoS选型","skillType":"3","callCount":7}]},{"name":"dingping","slug":"Kioh5Oipwq","avatarId":5,"tagline":"dingping 是一位将 AI 辅助开发深度融入日常工作流的医疗前端工程师，负责手术室终端设备的研发。TA 既能在业务系统中快速交付（患者资料管理、手术信息录入、多方会议协作），又能追踪 AI 系统架构的前沿话题（Agent 记忆系统、RAG 优化）。TA 与 AI 的协作极其高效——不是简单发号施令，而是通过精细化的参数控制和方法论引导，让 AI 成为可预测的生产力工具。快速原型、资产复用、标准化文档是 TA 的三大工作信条，体现出敏捷与工程化的平衡。","totalCalls":0,"totalTokens":21732848,"sessionsAnalyzed":243,"topDomains":["医疗前端开发（手术室终端）","AI辅助开发与工具集成","会议协作系统（实时画板）","技术文档与知识工程","AI Agent 架构（RAG与记忆系统）"],"roastTitle":"","projects":[],"gripHi":["前端开发（Electron桌面端）","医疗业务系统实现","知识管理与自动化"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"zd W","slug":"ON4ro9y5p6","avatarId":11,"tagline":"zd W 是一位在 Agent 开发、记忆压缩、多智能体系统设计领域拥有深厚造诣的专家。TA 既有底层算法优化的扎实功底，又有上层 Agent 系统架构的宏观视野；既能设计高效的记忆压缩算法，又能在生产环境中构建可扩展的多智能体协作系统。TA 特别擅长通过系统化的评估框架来验证 Agent 能力，展现出对智能系统鲁棒性的极致追求。TA 追求技术的本质，注重理论与实践的结合，是一位真正的「智能体系统架构师」。","totalCalls":0,"totalTokens":21039261,"sessionsAnalyzed":73,"topDomains":["Agent 系统架构设计","记忆压缩与检索优化","多智能体协作机制","大语言模型应用开发","分布式系统工程","提示工程与评估","数据处理与可视化"],"roastTitle":"","projects":[{"name":"层次化记忆压缩 Agent 平台"},{"name":"多智能体协作系统"},{"name":"提示工程与评估框架"}],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"yoang loo","slug":"OA6jVUWaeZ","avatarId":1,"tagline":"yoang loo是那种把 AI 工具当“生产系统”来驯化的人：先跑通，再拉满，再对标行业头部。和他协作时最明显的感受是节奏快、目标硬、容忍度低，但每一句催促背后都指向可落地的结果。你很少看到他停在抽象讨论里，他更在乎今天能不能真的多打通一条链路。","totalCalls":0,"totalTokens":20260728,"sessionsAnalyzed":43,"topDomains":["AI 工具编排与自动化","跨平台工作流打通","权限与集成调试","社交媒体内容生产","个人知识系统管理"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Elise Fournier","slug":"elise-fournier","avatarId":40,"tagline":"Model-structure-first. Before writing any code, she draws the compartmental diagram on paper — which boxes, which arrows, which parameters are known vs. estimated. Then translates the diagram to differential equations, then to Stan code. The biology drives the model, never the other way around.","totalCalls":30,"totalTokens":19800000,"sessionsAnalyzed":30,"topDomains":["Compartmental epidemic models (SIR/SEIR/SEIRS)","Bayesian model fitting with Stan","Real-time epidemic forecasting"],"roastTitle":"The Uncertainty Evangelist","projects":[],"gripHi":[],"gripLo":[],"quote":"Show me the R-hat and the traceplots before you interpret any parameter estimates. If the chains haven't converged, the posterior is meaningless","oneLiner":"She will give you a forecast, but she'll wrap it in so much honest uncertainty that you'll actually make a better decision.","activeDays30":0,"skills":[{"id":"08192afc-a581-4f49-ade0-4610cdc9aab1","title":"Bayesian SEIR model fitting with Stan","skillType":"1","callCount":30}]},{"name":"andreafy","slug":"WUXKH1n7Rr","avatarId":11,"tagline":"andreafy之前是某头部互联网公司的HRBP，现在自己出来带创业团队。但她跟Claude Code的对话记录看起来更像个不挂名的产品经理——自己搭了候选人调研脚本、面试题库生成器、人才盘点看板，还主导了全公司的AI协作能力评估。从容器创业公司recruiter到大厂BP再到现在自己出来带创业团队，她在技术团队里泡了几年后干了大多数HR不敢想的事：自己写工具替代自己的重复劳动。","totalCalls":0,"totalTokens":19488940,"sessionsAnalyzed":28,"topDomains":["用AI搭建HR内部工具","候选人调研与简历交叉验证","全公司AI协作能力评估","面试流程产品化","组织数据可视化"],"roastTitle":"","projects":[],"gripHi":["评估框架与考核标准","候选人调研方法","工具需求与产品设计"],"gripLo":["代码实现","前端样式"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"DabAZ","slug":"9RLcrZQZqq","avatarId":7,"tagline":"DabAZ comes across as a frontend-heavy practitioner who optimizes developer workflow and multilingual product quality in parallel. Their artifact trail favors configuration-first stabilization: align tooling, enforce formatting consistency, then iterate quickly on product-facing content. The style is pragmatic, detail-aware, and biased toward maintainability.","totalCalls":0,"totalTokens":18997541,"sessionsAnalyzed":92,"topDomains":["Frontend engineering","Localization and internationalization","TypeScript and React tooling","Developer productivity workflows"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Priya Kapoor","slug":"priya-kapoor","avatarId":21,"tagline":"Rule-first, then implementation. Reads the regulation, extracts the logic, formalizes it into rules, then builds the system. Every decision tree must be traceable back to a specific regulatory paragraph. Thinks in compliance workflows, not user stories.","totalCalls":12,"totalTokens":18500000,"sessionsAnalyzed":32,"topDomains":["AML/KYC compliance automation","Regulatory rule engines","Python / Django backend","Transaction monitoring systems"],"roastTitle":"The Regulation Compiler","projects":[],"gripHi":[],"gripLo":[],"quote":"Which regulation requires this? Give me the paragraph number.","oneLiner":"She reads 200-page regulatory documents and turns them into YAML configurations. She is more machine than human.","activeDays30":0,"skills":[{"id":"c685ac26-9681-43d3-b1e0-29c09bfa39d9","title":"Explainable Compliance Rule Engine Design","skillType":"1","callCount":12}]},{"name":"logan zheng","slug":"5-zhiQG2ms","avatarId":1,"tagline":"logan zheng是一位资深的AI工程化专家，拥有超过10年的前端开发经验。他不仅深度使用AI辅助开发工具，更通过AI完成的交互规模达到了惊人水平，累计消耗了超过1750万Token，特别是在d-code-learn-code项目上（超过1590万Token）。他专门从事AI辅助开发工具的深度整合与定制，在d-code项目群中构建了多个应用程序，包括钱包应用(d-code-wallet)、学习代码应用(d-code-learn-code)、面试样例(d-code-interview-example)等。他深度使用MCP协议，配置了43个MCP服务和465个MCP工具，拥有3627次AI编码交互记录，展现了AI优先的现代开发实践。","totalCalls":0,"totalTokens":17579570,"sessionsAnalyzed":52,"topDomains":["软件开发","前端工程","AI工具集成","MCP协议","系统架构","DevOps","AI工程化","项目管理"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"小鱼","slug":"xiaoyu","avatarId":6,"tagline":"从市场数据倒推。先看 BSR 排名变化趋势、Review 增速、关键词搜索量，再决定要不要进这个品类。不看供应商、不看成本——先确认有需求，再谈供应。","totalCalls":22,"totalTokens":16800000,"sessionsAnalyzed":26,"topDomains":["亚马逊选品数据分析","BSR 趋势监控与预测","竞品 Listing 变化追踪"],"roastTitle":"数据选品狙击手","projects":[],"gripHi":[],"gripLo":[],"quote":"这个 ASIN 的 BSR 最近 30 天从 5000 掉到 800，Review 增速也快，有人在推这个品","oneLiner":"白天做运营，晚上写爬虫（让 AI 写），每天早上看 5000 个 ASIN 的体检报告。","activeDays30":0,"skills":[{"id":"09dc6dc4-975d-4ba0-ac23-f198cc7bba00","title":"亚马逊 BSR 异常检测方法","skillType":"1","callCount":22}]},{"name":"小凡","slug":"xiaofan","avatarId":12,"tagline":"需求驱动，产品经理给什么就做什么，但会在实现层面做一些优化。不太纠结技术选型，够用就行。","totalCalls":6,"totalTokens":15600000,"sessionsAnalyzed":28,"topDomains":["微信小程序","支付宝小程序","Taro 跨端框架","小程序性能优化"],"roastTitle":"双平台搬砖人","projects":[],"gripHi":[],"gripLo":[],"quote":"这个 API 微信有但支付宝没有，得写个兼容层","oneLiner":"每天在两个小程序平台之间反复横跳，练就了一身跨端兼容的本事。","activeDays30":0,"skills":[{"id":"9c9a3c6a-7678-406d-8a6f-ee5ceca17ad5","title":"Taro 跨端兼容方案","skillType":"1","callCount":6}]},{"name":"Olga Ivanova","slug":"olga-ivanova","avatarId":4,"tagline":"Data-first. Starts by looking at actual defect images, understanding the failure modes, then choosing the simplest technique that reliably catches them. Never starts with the model.","totalCalls":20,"totalTokens":14200000,"sessionsAnalyzed":24,"topDomains":["Defect Detection (Surface, Dimensional, Assembly)","YOLO Object Detection & Training","Classical Image Processing (OpenCV)","Edge Deployment (Jetson, OpenVINO)"],"roastTitle":"The False Positive Hunter","projects":[],"gripHi":[],"gripLo":[],"quote":"Show me the reject images before we talk about models","oneLiner":"Will fight you if your 99% accurate model has a 3% false positive rate that shuts down the production line.","activeDays30":0,"skills":[{"id":"1caba017-3166-4bd4-a998-9b52d850848f","title":"Two-Stage Defect Detection Pipeline","skillType":"1","callCount":20}]},{"name":"D curry","slug":"m7yjgrjxY9","avatarId":0,"tagline":"D curry对 AI 的要求很像对同事的要求：别空谈，先把可交付的东西写出来。TA喜欢把需求钉死在‘文件、格式、进度回写’这种硬约束上，让工作能被接力。\nTA一边敢推翻旧链路（不介入平台式处理），一边又很现实地收敛选型（向量库、Embedding、老数据不背）。\n控制‘方向和交付物’，放手‘挖资料和敲代码’，这是 TA 跟 AI 合作的默认姿势。","totalCalls":0,"totalTokens":12455417,"sessionsAnalyzed":250,"topDomains":["工程文档与交付","知识库检索与向量系统","AI 代理服务工程","日志分析与可观测性","后端服务与接口设计"],"roastTitle":"","projects":[],"gripHi":["架构与技术选型（AI 链路、向量库、Embedding）","文档交付与上下文管理（交接文档、README、进度回写）"],"gripLo":["代码实现细节（按方案落地）"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Sarah Goldman","slug":"sarah-goldman","avatarId":18,"tagline":"Output-driven. Starts from what the MD needs to see in the pitch book, works backward to figure out what data she needs and how to get it. Thinks in deal timelines and deliverables, not sprints and tickets.","totalCalls":8,"totalTokens":12400000,"sessionsAnalyzed":22,"topDomains":["Financial modeling (DCF, LBO, comps)","Excel automation (VBA, openpyxl)","Python data processing","Pitch book data pipelines"],"roastTitle":"The Excel Survivor","projects":[],"gripHi":[],"gripLo":[],"quote":"Does the DCF tie out? I don't care how the code looks, just tell me the number matches.","oneLiner":"Her Python scripts have no tests, no types, and no documentation. They also produce pitch books that close billion-dollar deals.","activeDays30":0,"skills":[{"id":"c44b7cd6-4b87-41f0-821a-6aefda3bf2f2","title":"Bloomberg BQL Data Extraction for Comps","skillType":"3","callCount":8}]},{"name":"lvyxyz","slug":"nYZKdvdDTv","avatarId":4,"tagline":"lvyxyz 是一位专注于基准测试与架构设计的资深工程师，他们带着极致的工程鲁棒性和敏锐的产品感，正在重新定义评测工具的‘泛化高度’。他们不只是在编写代码，而是在通过 Memory Bank 和文档对齐，构建一套人机协同的高效骨架。","totalCalls":0,"totalTokens":11607090,"sessionsAnalyzed":20,"topDomains":["AI 内容安全与文本风控","软件基准测试与性能评估 (Benchmarking)","Electron 跨平台工具开发","高性能数据清洗 (CSV/JSON/Log Processing)"],"roastTitle":"","projects":[],"gripHi":["软件基准测试与性能评估 (Benchmarking)","Electron 跨平台工具开发","AI 内容安全与文本风控"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Rachel Cho","slug":"rachel-cho","avatarId":8,"tagline":"Starts from the design problem — what's tedious, error-prone, or inconsistent in the current workflow — and works backward to a scripting solution. Never automates for the sake of automation; every tool solves a real pain point.","totalCalls":8,"totalTokens":11500000,"sessionsAnalyzed":20,"topDomains":["Revit API / C# Plugin Development","Dynamo Visual Programming","BIM Automation & Model Quality"],"roastTitle":"The Architect Who Automates","projects":[],"gripHi":[],"gripLo":[],"quote":"That Dynamo graph looks like a plate of spaghetti. If you can't trace the logic from left to right, nobody else on the team will be able to either","oneLiner":"Started coding because she was tired of manually tagging 200 doors. Now she's her firm's unofficial CTO and the BIM coordinator finally sleeps through the night.","activeDays30":0,"skills":[{"id":"5fe06c06-f16f-4f35-baea-bc53e10e89ca","title":"Revit API Transaction & Error Handling Pattern","skillType":"3","callCount":8}]},{"name":"Yanhe Liu","slug":"WY0mqwRMaU","avatarId":11,"tagline":"他从自己的项目历史中提炼标准，而不是从外部借用规范——这说明他把真实代码库视为唯一可靠的知识来源。在动手之前，他先用CLAUDE.md建立AI的认知框架，让生成过程在规范的约束下展开。学习新技术时，他要求的不是最小示例，而是完整的思维地图：全面、有注释、有思路。当一个集成被证明可行，他会立刻让整个体系同步跟上——这是一种横向扩展的本能。行为特征显示他少说多做，是那种不下班的人。","totalCalls":0,"totalTokens":11149561,"sessionsAnalyzed":12,"topDomains":[],"roastTitle":"","projects":[],"gripHi":["frontend","ai-tooling"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Sam Brooks","slug":"sam-brooks","avatarId":8,"tagline":"Spatial-first: he visualizes the 3D scene in his head, then figures out the math to generate it. Code is a way to describe geometry procedurally rather than manually placing every vertex.","totalCalls":20,"totalTokens":10500000,"sessionsAnalyzed":19,"topDomains":["Blender 3D modeling & rendering","Python scripting for Blender (bpy)","Procedural geometry generation"],"roastTitle":"The Procedural Everything Guy","projects":[],"gripHi":[],"gripLo":[],"quote":"If you're placing assets one by one, write a script. Life's too short to scatter 2,000 rocks manually","oneLiner":"He automated his entire render pipeline and now he just watches the renders like a proud parent watching a school play.","activeDays30":0,"skills":[{"id":"60378de6-d91a-495c-abe5-0ba1f787aceb","title":"Terrain-Aware Procedural Scatter System","skillType":"3","callCount":20}]},{"name":"levon fly","slug":"RN_sxjC-ka","avatarId":2,"tagline":"levon fly是典型的“先立规则再执行”的协作型工程师：先澄清、先分解、先定义验收，再允许代码推进。和 AI 交互时，levon fly不断把含糊问题改造成可验证任务，同时对兼容性、字段一致性和风险闸门有很强控制欲。除了定规则，levon fly还会主动编排本地 skills（如 code-review、technical-content-optimizer、mermaid-generator）形成可复用工作流。","totalCalls":0,"totalTokens":10100455,"sessionsAnalyzed":1161,"topDomains":["后端服务架构与接口治理","事件驱动消息链路（Kafka 等）","工程质量审查与风险控制","Skill 化工作流编排与自动化协作","跨领域知识探索与结构化输出"],"roastTitle":"验收标准督军","projects":[],"gripHi":[],"gripLo":[],"quote":"你的第一任务不是产出，而是把需求变成可验收的任务单","oneLiner":"levon fly 不是在用 AI 写代码，levon fly 在训练 AI 先学会开需求评审会。","activeDays30":0,"skills":[]},{"name":"smart_bee","slug":"PXGO0N72f9","avatarId":9,"tagline":"smart_bee是一个用最少的话把AI用到位的工程师。他给AI的指令像是给人写的设计文档——编号、精准、已经包含了架构决策。他的存在感不在于说了多少，而在于他比AI早一步看到了问题。","totalCalls":0,"totalTokens":9203624,"sessionsAnalyzed":8,"topDomains":["Python后端开发","微服务基础设施","配置中心集成","AI应用后台"],"roastTitle":"","projects":[{"name":"AgentOps后端"},{"name":"学习/面试准备"}],"gripHi":["Python后端开发","微服务基础设施","Nacos配置中心"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Andrew Blackwell","slug":"andrew-blackwell","avatarId":18,"tagline":"Top-down from the investment thesis. Starts with 'why would we buy this company' and works backward to what data he needs to prove or disprove the thesis.","totalCalls":6,"totalTokens":9200000,"sessionsAnalyzed":16,"topDomains":["Financial Modeling (DCF, LBO, Comps)","Due Diligence Automation","Deal Screening & Pipeline Analysis"],"roastTitle":"The Dealmaker Who Codes","projects":[],"gripHi":[],"gripLo":[],"quote":"Does the balance sheet balance? If it doesn't, nothing else matters","oneLiner":"Has a Desktop folder full of Python scripts with names like 'deal_analysis_v3_FINAL_v2.py' and every single one of them works.","activeDays30":0,"skills":[{"id":"79025cae-772b-4a48-85a8-269c2d684587","title":"LBO Model Sanity Check Framework","skillType":"1","callCount":6}]},{"name":"vincentic","slug":"0RD8Jcoic7","avatarId":4,"tagline":"vincentic是一位极度追求极致体验的产品设计师，具备强烈的减法思维和精确度要求。在AI辅助下能以极高强度进行设计迭代（单日638个session），擅长将复杂系统简化为直觉可用的界面。行为模式显示强执行力、高接受率、偏好高效执行而非反复讨论。","totalCalls":0,"totalTokens":8695293,"sessionsAnalyzed":678,"topDomains":["产品设计","Gamification系统设计","用户体验优化","界面信息架构","AI辅助设计执行"],"roastTitle":"","projects":[{"name":"Gamified Productivity Platform（ gamified productivity 应用设计）"}],"gripHi":["Product Design & UI/UX","Gamification 系统设计","AI Tool Orchestration"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Ava Chen","slug":"ava-chen","avatarId":0,"tagline":"Learn by doing, then understand why. Ava Chen builds first to get something working, then goes back to understand the underlying pattern — and she's getting faster at closing that gap.","totalCalls":14,"totalTokens":8500000,"sessionsAnalyzed":18,"topDomains":["Vue.js Components & Composition API","TypeScript Fundamentals","CSS Layout (Flexbox/Grid)"],"roastTitle":"The Sponge","projects":[],"gripHi":[],"gripLo":[],"quote":"wait, why does this work?","oneLiner":"Asked 'but why does this work?' so many times that the AI started explaining things preemptively.","activeDays30":0,"skills":[{"id":"0fb5b4fe-2e63-450b-84d9-b7c2cb94dd55","title":"Vue Composition API Patterns","skillType":"1","callCount":8},{"id":"d45a76ba-b76b-40d0-ac45-9444947d89de","title":"TypeScript Basics for Vue Components","skillType":"1","callCount":6}]},{"name":"冷爽（Evan）","slug":"YeAXZjUzbK","avatarId":7,"tagline":"冷爽（Evan） 是那种先把页面在脑子里画清楚，再让 AI 补代码的人。TA 对活动页里每一块的布局、背景色和信息密度都有强烈的统一追求，会一遍遍用“和上面那个板块保持一致”“模块底色为 #f4f4f4”之类的话，把页面塑造成心里早已有的样子。同时，冷爽（Evan）对数据契约和上线链路也抓得很紧，既不接受前端乱兜底，也不把构建部署当成“黑盒脚本”。","totalCalls":0,"totalTokens":8445192,"sessionsAnalyzed":6,"topDomains":["Vue 技术栈的活动/专题落地页开发","前端工程化与构建部署流水线","面向用户的文案与分享体验设计"],"roastTitle":"","projects":[{"name":"Arm 主题活动页（匿名化）"}],"gripHi":["页面布局与模块风格统一","数据字段与兜底策略"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"ACL J","slug":"ZU0ukgGGsb","avatarId":0,"tagline":"ACL J 是那种会先把“这个系统到底该成为什么样”写出来，再决定代码怎么铺的人。他对产品语义、系统边界和对外表达都抓得很紧，尤其讨厌拿术语、兼容性和零散补丁来掩盖真正没想清楚的问题。和 AI 协作时，他最常做的不是催代码，而是重设坐标系。","totalCalls":0,"totalTokens":8271020,"sessionsAnalyzed":57,"topDomains":["检索增强生成与知识库产品","Go/React 全栈系统设计","后台产品化与信息架构","可靠性治理与接口语义","浏览器自动化与流程脚本"],"roastTitle":"","projects":[{"name":"个人书籍知识库问答平台"},{"name":"RAG 评测与运维控制台"}],"gripHi":["产品定义与信息架构","架构边界与技术取舍","可靠性语义"],"gripLo":["大块执行铺设"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"王志明","slug":"NYv_Ze_KqB","avatarId":8,"tagline":"王志明是一位以「先诊断再下刀」为本能的前端工程师，在支付系统和认证模块这两块高stakes的领域里摸爬滚打，有一套属于自己的代码组织世界观——业务语义优先、边界不能扩散、utils文件是垃圾桶。他用AI的方式有点反直觉：先要求分析，不让改；改了不好就回滚；AI上下文找错了，他比AI更早发现。","totalCalls":0,"totalTokens":7824991,"sessionsAnalyzed":10,"topDomains":["支付系统集成与调试","前端工程化与组件迁移","登录/认证系统","React/Next.js性能优化","UI设计还原"],"roastTitle":"","projects":[],"gripHi":["支付系统","前端工程化","认证架构"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"James Whitfield","slug":"james-whitfield","avatarId":16,"tagline":"Starts with the investment thesis — what kind of asset, what market, what's the value-add opportunity — then builds the analysis to prove or disprove it. Numbers serve the thesis, not the other way around.","totalCalls":8,"totalTokens":7800000,"sessionsAnalyzed":15,"topDomains":["Commercial Real Estate Underwriting","Rent Roll Analysis & Processing","Cap Rate & NOI Calculations"],"roastTitle":"The Rent Roll Whisperer","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the in-place cap rate vs. the stabilized cap rate? If the spread isn't at least 150 bps, the value-add isn't worth the execution risk","oneLiner":"Can spot a suspicious expense ratio from across the room and has a Python script that processes rent rolls faster than his broker can send them.","activeDays30":0,"skills":[{"id":"1f1a2d07-d7a8-449f-a6e4-8a1a32eedd49","title":"Rent Roll Processing & Mark-to-Market Framework","skillType":"3","callCount":8}]},{"name":"Demon X (Demon_by)","slug":"QK8Exxns6y","avatarId":7,"tagline":"Demon X (Demon_by) 给 AI 的指令看起来像任务分配，但底层其实是在做系统建模：先定边界、再排顺序、最后才允许实现。TA 对“文档先行”和“减法清理”有很强执念，愿意延后短期进度来换长期可维护性。和 Demon X (Demon_by) 协作时，你会明显感觉到：TA 不是在要一个答案，而是在训练一个能长期工作的工程脑。","totalCalls":0,"totalTokens":7192810,"sessionsAnalyzed":349,"topDomains":["回合制/卡牌玩法系统设计","前后端任务链路与事件契约治理","工程文档驱动开发流程","状态机与资源系统建模","技术债清理与可维护性优化"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Pengfei Li","slug":"3kNkL4xw5I","avatarId":11,"tagline":"Pengfei Li像那种会把模糊方向硬生生压成执行协议的人。TA 不太接受“先做个 demo”或“差不多能跑”，会一路追问到边界、失败场景、上板闭环和交付格式都讲清楚。跟 AI 协作时，TA 最常做的不是下命令，而是校准标准：要一步步讲、要解耦、要贴真实环境、要最后拿出真正结果。","totalCalls":0,"totalTokens":7157807,"sessionsAnalyzed":5,"topDomains":["FPGA 架构与板级部署","脉冲神经网络 / Transformer 协同优化","高并发票务与交易一致性","Java RPC / 服务接口设计","实验协议与技术交付"],"roastTitle":"","projects":[],"gripHi":["系统目标与验收口径","架构边界与耦合度","解释顺序与学习体验"],"gripLo":["脚本与批处理落地"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Xinye Liao","slug":"Sp1VsTIn3-","avatarId":0,"tagline":"Xinye Liao 像那种会让 AI 一直返工到“既对、又准、又不废话”的人。题目跨度很大，从少样本目标检测论文、RLHF 公式、网络容量评估到业务面试准备，但他抓得最紧的始终是同一件事：别瞎编，别漏关键，别拿漂亮句子掩盖没想清楚的地方。AI 在他这里可以是起草员、陪练、讲师，唯独不能代替他对技术边界和表达质量的最后裁决。","totalCalls":0,"totalTokens":7005357,"sessionsAnalyzed":7,"topDomains":["少样本目标检测与计算机视觉研究","技术论文写作与实验叙事打磨","大模型后训练与推理机制理解","网络流量分析与容量决策","AI 业务方案与面试准备"],"roastTitle":"","projects":[{"name":"少样本目标检测毕业论文"},{"name":"链路容量评估与扩缩容决策"},{"name":"大模型与算法面试准备"}],"gripHi":["技术表述与措辞边界","方法完整性与结构闭环","分析假设与决策理由"],"gripLo":["大段讲解与初稿铺陈"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"吴一鸣","slug":"P-2yQcY58b","avatarId":11,"tagline":"吴一鸣像一个把工程当产品来做的人：写代码时盯机制边界，做协作时盯流程可复用。TA不会满足于“修好了”，而是不断追问这个解法是否优雅、是否可验证、是否能被团队长期使用。和 AI 配合时，TA更像技术负责人在带设计评审——不断重定义问题、收敛接口、压实验收标准。","totalCalls":0,"totalTokens":6332361,"sessionsAnalyzed":44,"topDomains":["跨端工程规范","前端性能与加载链路","组件库治理","复杂交互调试","AI工作流设计"],"roastTitle":"","projects":[],"gripHi":["规则边界定义","交互修复验收"],"gripLo":["文档润色与格式化"],"quote":"","oneLiner":"","activeDays30":9,"skills":[]},{"name":"胡建华","slug":"1ZSyFPH4Nr","avatarId":6,"tagline":"胡建华是一位学习导向的全栈开发者，正在使用多种AI工具辅助开发简历生成器项目。TA擅长系统化调试，不仅解决问题还要理解问题本质。","totalCalls":0,"totalTokens":5862412,"sessionsAnalyzed":7,"topDomains":["前端开发","后端开发","数据库","AI辅助开发"],"roastTitle":"","projects":[],"gripHi":["前端开发","后端开发","数据库"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":2,"skills":[]},{"name":"Stove3 Wang","slug":"v72j2VQ1Xa","avatarId":3,"tagline":"Stove 是一个对失控感极度敏感的人。他不要 AI 闷头干活，他要全程透明；他不要 AI 迎合，他要从第三方视角审视真相。","totalCalls":0,"totalTokens":5836439,"sessionsAnalyzed":18,"topDomains":["即时零售/电商","AI Agent 协作","商业分析（ROI/周转率/杜邦分析）","跨境出海（沙特市场）"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"freeman","slug":"Zy8vHS97oS","avatarId":6,"tagline":"freeman更像是在用 AI 做“共同推理”：先把边界条件说清楚，再让实现细节自动展开。TA 会频繁纠偏，把讨论从“写点代码试试”拉回到“这件事在哪一层、有什么硬约束”。","totalCalls":0,"totalTokens":4849386,"sessionsAnalyzed":10,"topDomains":["交互与界面设计","工程决策与边界","配置与部署","Agent 架构与工具链"],"roastTitle":"","projects":[{"name":"本地学习/练习项目"},{"name":"本地 Agent/工具项目"},{"name":"本地测试/实验项目"}],"gripHi":["边界条件与分层纪律","配置/密钥真值"],"gripLo":["实现细节（让 AI 去填）"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"huanglvjing","slug":"BWXCd0mldy","avatarId":2,"tagline":"huanglvjing 是一位能够精准驾驭 AI 的全栈 + 嵌入式开发者。TA 不只是让 AI 写代码，而是持续引导 AI 走向更简洁、更准确、更符合工程判断的方向；同时用真实数据验证 AI 输出，用追问原理的方式将 AI 转化为学习加速器。技术栈横跨 Web 前后端、STM32 嵌入式和 CLI 工具，在医疗、校园、金融科技、物联网等多个领域均有完整项目交付。","totalCalls":0,"totalTokens":4557691,"sessionsAnalyzed":47,"topDomains":["全栈Web开发","嵌入式物联网","AI协作与驾驭","金融科技平台","医疗看护系统"],"roastTitle":"","projects":[],"gripHi":["AI协作与驾驭","全栈Web开发","嵌入式物联网"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Patrick Chen","slug":"patrick-chen","avatarId":3,"tagline":"Top-down, hypothesis-first. Starts with 'what do we need to prove?' then works backward to the data. Thinks in 2x2 matrices and issue trees.","totalCalls":6,"totalTokens":4500000,"sessionsAnalyzed":15,"topDomains":["Market sizing & TAM/SAM/SOM","Client presentation design","Data analysis (Python/Excel)"],"roastTitle":"The Prompt Partner","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the 'so what' here? I can't put a data table on a slide without a takeaway","oneLiner":"Treats AI like a first-year analyst — clear briefs, tight deadlines, and a red pen that never runs dry.","activeDays30":0,"skills":[{"id":"609c7aad-e1f1-4411-bf7d-59a46bfdb092","title":"TAM/SAM/SOM Market Sizing with AI","skillType":"3","callCount":6}]},{"name":"Yajie hu","slug":"DdCsA4jw8B","avatarId":11,"tagline":"一名追求极致效率的架构师与系统开发者。习惯于将复杂的业务逻辑拆解为极简的结构化指令，把 AI 视为执行工作流的机械臂。通过建立高容错率的系统级数据契约与绝对的进度优先原则，擅长打破底层限制，实现人机混合工作流的无缝协同与效率最大化。","totalCalls":0,"totalTokens":4285600,"sessionsAnalyzed":842,"topDomains":["数据清洗与结构化","自动化工作流设计","工具链整合"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Samantha Cruz","slug":"samantha-cruz","avatarId":5,"tagline":"Regulation-first. Reads the actual regulation text, identifies the technical obligations, then designs the system to satisfy them. Doesn't trust summaries — reads Article 17 herself before building the right-to-erasure handler.","totalCalls":5,"totalTokens":4100000,"sessionsAnalyzed":14,"topDomains":["GDPR / CCPA Compliance Engineering","Data Subject Access Requests (DSAR)","Data Mapping & Classification"],"roastTitle":"The Deletion Enforcer","projects":[],"gripHi":[],"gripLo":[],"quote":"Soft-delete is not erasure. Article 17 says erasure. If the data is still on disk, it's not erased.","oneLiner":"Reads GDPR articles for fun and then writes code to make sure every system actually does what Article 17 says.","activeDays30":0,"skills":[{"id":"398fed54-a0d7-4bf6-be0e-294839cf6926","title":"Multi-System DSAR Pipeline Architecture","skillType":"3","callCount":5}]},{"name":"Ben Taylor","slug":"ben-taylor","avatarId":3,"tagline":"Systematic: start with the crawl data, identify patterns, prioritize by traffic impact, then automate the fix. Every SEO recommendation has a projected traffic number attached.","totalCalls":4,"totalTokens":3800000,"sessionsAnalyzed":12,"topDomains":["Technical SEO auditing at scale","Python web scraping & crawl analysis","Schema markup & structured data"],"roastTitle":"The Crawl Budget Accountant","projects":[],"gripHi":[],"gripLo":[],"quote":"Check the log files — I want to know what Googlebot is actually crawling, not what we think it should crawl","oneLiner":"He's read more nginx logs than actual web pages, and he's fine with that.","activeDays30":0,"skills":[{"id":"709ce05c-9ac1-4e1b-843e-c0e058f07744","title":"Googlebot Crawl Budget Audit from Log Files","skillType":"3","callCount":4}]},{"name":"XinweiMax","slug":"PW2iv2h-DR","avatarId":0,"tagline":"XinweiMax像带项目评审习惯的同学在跟 AI 结对：开口先看分层与测试顺序，把手写代码前先写清「为什么」。一旦察觉问题出在 JRE 或中文路径这种环境里，会立刻把叙事拽回业务失败本身，不让助手用环境借口糊弄过去。","totalCalls":0,"totalTokens":3644227,"sessionsAnalyzed":12,"topDomains":["Java/Spring 服务端与分层架构","测试驱动开发与领域建模","全栈拼装与本地工具链排障","微服务示例与认证/中间件集成","校招级作业与契约/集成测试"],"roastTitle":"","projects":[],"gripHi":["分层与交付节奏","构建与测试执行","问题归因与环境排障"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Sophie Martin","slug":"sophie-martin","avatarId":12,"tagline":"Story-first. Starts with 'what's the human impact?' then works backward to find the data that proves it. Every chart must answer a question a reader would actually ask.","totalCalls":4,"totalTokens":3600000,"sessionsAnalyzed":13,"topDomains":["Data scraping & cleaning","D3.js interactive visualization","Python data analysis"],"roastTitle":"The CSV Whisperer","projects":[],"gripHi":[],"gripLo":[],"quote":"If I can't source this number to a public record, it doesn't go in the article","oneLiner":"Will spend a week verifying numbers that AI generated in 5 seconds, because a published correction is a journalist's scarlet letter.","activeDays30":0,"skills":[{"id":"0c885766-2209-46df-bc59-515318034d8c","title":"Multi-State Web Scraping for Public Data","skillType":"3","callCount":4}]},{"name":"cjhzzx","slug":"zDPSa0H4Zt","avatarId":10,"tagline":"cjhzzx 是 Vue/React/TypeScript 全栈开发者，同时使用 Go 做后端。TA 最突出的特点是**规格驱动**——每次都先定义完整的产品规格和技术方案，而不是边做边改。TA 对视觉质量有明确的高标准，即使是小改动也会要求代码层面的优化。","totalCalls":0,"totalTokens":3310136,"sessionsAnalyzed":74,"topDomains":["前端开发 (Vue/React)","3D/WebGL 开发","全栈开发","UI/UX 设计实现","项目架构设计"],"roastTitle":"","projects":[{"name":"AmazingZoom"},{"name":"BugManagePlatform"},{"name":"KesouSearch"}],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Diana Costa","slug":"diana-costa","avatarId":14,"tagline":"Starts from the business metric (CAC, LTV, conversion rate) and works backward to what data she needs. Code is a means to an answer, never the point itself.","totalCalls":4,"totalTokens":3200000,"sessionsAnalyzed":12,"topDomains":["Event tracking & analytics instrumentation","SQL-based cohort analysis","A/B test design & statistical significance"],"roastTitle":"The Funnel Whisperer","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the event name for that? If it's not tracked, it didn't happen","oneLiner":"She'll instrument your grandmother's kitchen if she suspects there's a conversion event hiding in the pantry.","activeDays30":0,"skills":[{"id":"5d4b6c6f-3556-4c58-ae53-457098a0921a","title":"Multi-Touch Attribution Pipeline","skillType":"3","callCount":4}]},{"name":"大官人刘","slug":"LxUht87nOM","avatarId":7,"tagline":"大官人刘 是一位有强烈审美主张的系统架构师，正在构建一套 AI Agent 驱动的可视化大屏平台。TA 的口头禅是'从根本出发'——无论是一个坐标错误还是一套 JSON 解析方案，TA 都会追溯到架构层面解决，而不是就地打补丁。TA 对 AI 的态度像一个严格但耐心的导师：既会说'丑到爆了'，也会用图层叠加的类比帮 AI 建立正确的心智模型。","totalCalls":0,"totalTokens":2800000,"sessionsAnalyzed":210,"topDomains":["AI Agent 系统架构","可视化大屏 / 低代码平台","前端工程（Vue.js 组件化）","LLM 工程（Prompt 优化、结构化输出）","产品设计与 UI 审美"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Jasmine Wu","slug":"jasmine-wu","avatarId":11,"tagline":"User-first, hypothesis-driven. Starts every decision with 'what does the user actually need?' then validates with data. Thinks in user journeys, not feature lists.","totalCalls":5,"totalTokens":2800000,"sessionsAnalyzed":11,"topDomains":["Product requirements & PRDs","User research synthesis","Data analysis (SQL, basic Python)"],"roastTitle":"The User Story Evangelist","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the user story? I can't prioritize a feature without knowing who it's for","oneLiner":"Will not look at a feature proposal until you can name the user, state the problem, and show her the funnel drop-off chart.","activeDays30":0,"skills":[{"id":"0142dc5e-a1b4-45bf-a92c-9927929a6fbe","title":"AI-Assisted PRD Writing Framework","skillType":"3","callCount":5}]},{"name":"siqi ma","slug":"siqi-ma","avatarId":8,"tagline":"先看文献，再定假设，再设计实验。严格遵循研究范式，数据分析之前必须预注册分析方案。","totalCalls":3,"totalTokens":2400000,"sessionsAnalyzed":10,"topDomains":["教育数据挖掘(EDM)","学习分析(Learning Analytics)","Python/R 统计分析"],"roastTitle":"p值警察","projects":[],"gripHi":[],"gripLo":[],"quote":"你这个p值是0.048，要报告效应量和置信区间，不能只说显著","oneLiner":"在她的世界里，不报效应量和置信区间的分析结果跟不存在一样。","activeDays30":0,"skills":[{"id":"e1a88233-f12f-40b0-a4af-f8c4a2a4fa21","title":"理论驱动的教育数据特征工程","skillType":"1","callCount":3}]},{"name":"ckvv","slug":"9mggDzuCBC","avatarId":5,"tagline":"ckvv像那种会先把系统边界画干净，再开始往里填内容的人。TA 对 AI 的态度不是“帮我做完”，而是“先按我的判断方式来想”；一旦发现抽象太重、表达太虚、边界太脏，就会立刻收紧。你会觉得 TA 不只是想把东西做出来，而是想把结构、用户感知和协作认知一起做顺。","totalCalls":0,"totalTokens":2205065,"sessionsAnalyzed":249,"topDomains":["AI 产品前端与多端应用","Electron 与 Web 共构","知识库与智能问答界面","企业后台与运营平台","前端工程化与协作规范"],"roastTitle":"","projects":[{"name":"多端 AI 学习系统"},{"name":"企业知识问答产品"},{"name":"多模态搜索与内容生成产品"}],"gripHi":["信息架构与复杂度边界","架构规则与协作文档"],"gripLo":["AI 初稿生成"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Little Red Riding Hood","slug":"mPSj1NSXcW","avatarId":3,"tagline":"Little Red Riding Hood是7年经验的前端开发，擅长React+Vue+TypeScript，习惯少说多做。与AI协作时消息极简（平均38字符），倾向于先出产品原型再开发，决策果断。对工程化细节要求高——端口要区分、目录要扁平、依赖要清晰。技术好奇心强，会深挖闭包等代码原理，还自己创建Vue迁移Skill做知识沉淀。","totalCalls":0,"totalTokens":2170171,"sessionsAnalyzed":9,"topDomains":["前端开发","浏览器扩展","H5移动端","工程化","AI应用开发"],"roastTitle":"","projects":[{"name":"ai-translate"},{"name":"手串DIY+H5"},{"name":"examples-webpack"}],"gripHi":["前端开发","工程化"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Jasmine Okonkwo","slug":"jasmine-okonkwo","avatarId":3,"tagline":"Product-first. Starts with the user story, then figures out the technical implementation. Comfort with ambiguity — ships MVPs and iterates.","totalCalls":0,"totalTokens":2100000,"sessionsAnalyzed":10,"topDomains":["EdTech product development","React frontend","Adaptive learning algorithms"],"roastTitle":"The Teacher Who Ships","projects":[],"gripHi":[],"gripLo":[],"quote":"Does a teacher actually need this, or are we building it because it's cool?","oneLiner":"She went from grading homework to grading pull requests, and she applies the same rubric to both: does it actually help someone learn?","activeDays30":0,"skills":[]},{"name":"Ingrid Larsen","slug":"ingrid-larsen","avatarId":16,"tagline":"Systems-thinking applied to human processes. Sees every team dysfunction as a system problem, not a people problem. Draws feedback loops on whiteboards and process diagrams in Mermaid.","totalCalls":4,"totalTokens":1950000,"sessionsAnalyzed":9,"topDomains":["Agile process design & coaching","Jira API automation","Python scripting for team metrics"],"roastTitle":"The Process Hacker","projects":[],"gripHi":[],"gripLo":[],"quote":"If you're doing it manually more than twice, it should be automated","oneLiner":"An agile coach who can actually code — which makes her either the most useful coach you've had or the most dangerous, depending on your Jira hygiene.","activeDays30":0,"skills":[{"id":"07fed5a5-b862-4664-bc6f-6b82edb088b3","title":"Jira API Sprint Metrics Automation","skillType":"3","callCount":4}]},{"name":"zz","slug":"MKKo8fSFzs","avatarId":6,"tagline":"zz像那种会一边研究 AI 工具，一边琢磨怎么把它包装成能卖、能交付、能跑通结果的人。TA不太迷恋抽象讨论，更相信先把东西做出来，再看这件事值不值得继续加码。面对模糊需求时，TA 的第一反应不是兴奋，而是先把边界划清楚。","totalCalls":0,"totalTokens":1722422,"sessionsAnalyzed":6,"topDomains":["AI 工具安装与本地配置","页面复刻与 Figma 落地","接单流程判断与交付包装","本地评测与结果分享"],"roastTitle":"","projects":[],"gripHi":["任务范围与交付边界","商品化包装"],"gripLo":["具体实现细节"],"quote":"","oneLiner":"","activeDays30":1,"skills":[]},{"name":"Jordan Ellis","slug":"jordan-ellis","avatarId":12,"tagline":"Process-first. Maps out the workflow before writing any code. Who needs what, when, and in what format? The tool comes after the process is clear.","totalCalls":4,"totalTokens":1680000,"sessionsAnalyzed":7,"topDomains":["Legal Workflow Automation","Contract Lifecycle Management","Document Template Generation"],"roastTitle":"The Plumber of Legal","projects":[],"gripHi":[],"gripLo":[],"quote":"What happens if they leave the entity name blank? Because they will.","oneLiner":"Reduced NDA turnaround from 3 days to 4 hours, and nobody even noticed because that's what good plumbing looks like.","activeDays30":0,"skills":[{"id":"f1949ab7-a28a-42f8-ab3c-ead5e45a2d7c","title":"Legal Intake Triage Automation","skillType":"3","callCount":4}]},{"name":"马学冬","slug":"1JyvQf_QnX","avatarId":0,"tagline":"马学冬 是一名全栈型 AI 应用开发工程师，既能深入底层基础设施（VPC 路由、消息队列持久化），又能驾驭上层 AI 应用（RAG 知识库、多模态微调）。TA 的典型特征是不满足于'能用'，总要追问'为什么这样设计'——从 IAM 信任策略到 SSL 证书验证，从分布式限流到 Publisher Confirm，每一步都要理解本质。这种打破砂锅问到底的劲头，让 TA 在架构师和工程师两种角色间自如切换。","totalCalls":0,"totalTokens":1671985,"sessionsAnalyzed":15,"topDomains":["AI 应用开发（RAG、大模型微调、多模态识别）","后端架构（消息队列、分布式限流、高并发系统）","云基础设施（AWS EC2/RDS/VPC/ALB/IAM）","知识库系统（PDF 解析、向量检索、混合检索）","Java/Spring Boot 企业级开发"],"roastTitle":"","projects":[{"name":"RAG 知识库系统"},{"name":"消息发送引擎"},{"name":"订单识别系统"}],"gripHi":["原理理解","代码实现"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Elena Volkov","slug":"elena-volkov","avatarId":8,"tagline":"Graph-based. Thinks in nodes and branches. Every conversation is a tree, every player choice is an edge, every character has a state. Writes dialogue in her head but implements it as logic.","totalCalls":4,"totalTokens":1450000,"sessionsAnalyzed":8,"topDomains":["Branching dialogue systems","Ink scripting language","Character voice & tone"],"roastTitle":"The Voice Police","projects":[],"gripHi":[],"gripLo":[],"quote":"If the player can't tell that their choice mattered, then the choice didn't matter","oneLiner":"Will reject your AI-generated dialogue, rewrite it in front of you, and explain why the character would never use the word 'facilitate.'","activeDays30":0,"skills":[{"id":"f23b1a4f-8813-46a1-a991-4d926075fa4e","title":"Ink Branching Dialogue Architecture","skillType":"1","callCount":4}]},{"name":"Laura Bianchi","slug":"laura-bianchi","avatarId":18,"tagline":"Spatial-first. Always starts by asking 'what does this look like on a map?' before diving into numbers. Thinks in layers — demographic data, transit networks, land use, all overlaid.","totalCalls":5,"totalTokens":1350000,"sessionsAnalyzed":8,"topDomains":["GIS data processing & spatial analysis","Interactive map visualization (Mapbox GL JS)","Transit accessibility analysis"],"roastTitle":"The Cartographic Policy Wonk","projects":[],"gripHi":[],"gripLo":[],"quote":"Wait, what CRS is this shapefile in? If it's NAD83 and you're mixing it with WGS84 we're going to have offset issues","oneLiner":"She will reject your beautiful map if the CRS is wrong, the legend is missing, or the color scale hides the pattern that matters.","activeDays30":0,"skills":[{"id":"d38ebbf7-d76b-4750-bc0a-9638372f0104","title":"Transit accessibility isochrone analysis","skillType":"1","callCount":5}]},{"name":"Yang Yang","slug":"T00tTfPst7","avatarId":4,"tagline":"Yang Yang像一个“会给 AI 立规矩”的开发者：先把问题定性，再决定怎么做。TA 频繁用边界、职责、以及审美一致性来筛选方案，让对话从‘能不能跑’走向‘值不值得这么做’。","totalCalls":0,"totalTokens":1314667,"sessionsAnalyzed":31,"topDomains":["前端/网页实现","测试与工程化约束","脚本与数据处理","AI 工具协作与工作流"],"roastTitle":"","projects":[{"name":"匿名项目（基于近 30 天对话）"}],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Gao12123","slug":"u9CADCuM0B","avatarId":1,"tagline":"Gao12123 是一位具有卓越架构设计能力的前端工程师。在财务管理系统的开发中，通过系统化的架构优化、流程重设和规范建立，实现了代码质量和系统可维护性的显著提升。展现出了从问题诊断到架构级别优化的完整工程化思维。","totalCalls":0,"totalTokens":1250000,"sessionsAnalyzed":232,"topDomains":["前端架构设计与优化","文件上传流程架构","API 规范化与系统分层","组件架构与职责划分","数据流管理与优化"],"roastTitle":"","projects":[],"gripHi":["前端架构设计与优化","系统分层与职责划分","API 规范化与工程化"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"梁旭","slug":"liang-xu","avatarId":9,"tagline":"从监管要求和精算准则出发，先确定计算口径和假设，然后逐步构建模型。每一步都要跟 Excel 旧模型交叉验证。","totalCalls":5,"totalTokens":1200000,"sessionsAnalyzed":7,"topDomains":["寿险精算模型（准备金、定价、利润测试）","偿二代（C-ROSS）资本计算","Python 精算建模"],"roastTitle":"精算界的 Python 先驱","projects":[],"gripHi":[],"gripLo":[],"quote":"这个准备金的计算结果跟 Excel 差了 0.3%，你帮我查一下是哪一步的贴现因子不一样","oneLiner":"把 47 个 sheet 的 Excel 模型重写成 Python，然后花了比写代码更长的时间来验证两边结果一致。","activeDays30":0,"skills":[{"id":"0926b87a-cf8e-4a94-a7fe-868ef84092cb","title":"精算模型 Excel 到 Python 迁移方法论","skillType":"3","callCount":5}]},{"name":"阿翔","slug":"axiang","avatarId":6,"tagline":"数学先行。先在纸上推公式，再在Excel里建模型，最后用Python跑模拟验证。不相信直觉，只相信数据。每一个掉落概率、每一条经验曲线都要经过至少10万次模拟。","totalCalls":4,"totalTokens":980000,"sessionsAnalyzed":6,"topDomains":["游戏数值平衡设计","掉落概率与随机系统","蒙特卡洛模拟"],"roastTitle":"概率论原教旨主义者","projects":[],"gripHi":[],"gripLo":[],"quote":"这个掉落概率你是怎么定的？给我看推导过程","oneLiner":"一个不会写好看代码但能在纸上推导保底公式的数值策划，让程序员和策划都怕他。","activeDays30":0,"skills":[{"id":"d1253049-6684-49f3-9a11-60ab2a7dd168","title":"抽卡保底机制数学设计","skillType":"1","callCount":4}]},{"name":"lina","slug":"y1tP84sue7","avatarId":0,"tagline":"lina像一位把系统当值班体系来做的人：先把触发条件、检查节拍、消息链路钉死，再让AI去执行。TA不追求纸面完美，而是追求“现在就能跑、出了事能第一时间到人”。和TA协作时会明显感到，结果闭环比过程华丽更重要。","totalCalls":0,"totalTokens":970499,"sessionsAnalyzed":28,"topDomains":["事件监控自动化","金融与地缘新闻追踪","告警策略与通知编排","任务调度与运行窗口设计"],"roastTitle":"","projects":[{"name":"实时事件与市场波动监控系统"},{"name":"聊天入口消息代理与调度"}],"gripHi":["告警触发与通知链路","运行节拍与时间窗口"],"gripLo":["实现细节执行"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"jiahui liu","slug":"_ySxzG0BFK","avatarId":11,"tagline":"jiahui liu是一位'大模型优先'的AI系统工程师，在多智能体架构领域有深厚积累。TA的思维方式是：遇到复杂问题先考虑能否用大模型解决，而非编写规则。在系统设计上，TA对职责边界有近乎偏执的清晰度——数据获取和数据分析必须分离，日志和代码是两种不同的真相。","totalCalls":0,"totalTokens":951278,"sessionsAnalyzed":166,"topDomains":["多智能体系统架构","LLM应用开发","数据智能与Text2SQL","零售业务分析","RAG与知识检索","Agent评测系统","提示词工程"],"roastTitle":"","projects":[{"name":"零售业务多智能体对话系统"},{"name":"商品智能匹配系统"},{"name":"Text2SQL对话系统"}],"gripHi":["多智能体系统架构","LLM应用开发","数据智能与Text2SQL"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"michael","slug":"hIFTdacEsJ","avatarId":10,"tagline":"michael 在 gRPC、网关与前端这条线上既问原理又追落地，会先设边界再要示例，对「做完」和「做对」分得很清，会主动要求用 shell 等手段验证。偏结果导向、可验证、功能完整对齐。","totalCalls":0,"totalTokens":866495,"sessionsAnalyzed":23,"topDomains":["gRPC/流式RPC","API 网关与代理","Web 前端与协议转换","构建与进程管理","浏览器与后端通信"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":["gRPC/流式RPC"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"刘帅彬","slug":"fI6ocoTFEY","avatarId":0,"tagline":"刘帅彬 一边搭生物医学文献方向的知识图谱原型（PDF→实体关系→图库），一边做实用的文献格式小工具；协作时喜欢用终端与日志原文缩小环境问题，让助手少猜、多对症。","totalCalls":0,"totalTokens":842357,"sessionsAnalyzed":3,"topDomains":["知识图谱与文献信息抽取","Python 数据工程与可视化界面","Web 前端与第三方 API 集成","开发者工具链与本地环境（Windows）","学术写作与参考文献规范"],"roastTitle":"","projects":[],"gripHi":["知识图谱与文献信息抽取","Python 工程与 Streamlit","Web 前端与 API"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"lizhenlong","slug":"y9uVEOUFVI","avatarId":11,"tagline":"lizhenlong是结果导向的业务工程型开发者，擅长在复杂后台场景里同时处理性能、体验和规则边界。TA与AI协作不是“交给AI做完”，而是把AI纳入自己的验证循环：给约束、看结果、再纠偏。整体风格是快速推进但不牺牲一致性，短期交付和长期可维护都会兼顾。","totalCalls":0,"totalTokens":829594,"sessionsAnalyzed":23,"topDomains":["前端工程与Vue生态","保险结算与佣金配置系统","接口联调与数据建模","登录鉴权与路由治理","页面性能与交互体验优化"],"roastTitle":"","projects":[],"gripHi":["登录鉴权与路由治理","列表性能与交互一致性"],"gripLo":["通用脚本与环境操作"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"大伟","slug":"dawei","avatarId":12,"tagline":"从项目整体的投资逻辑出发——地块什么条件、产品怎么排布、售价怎么定、去化节奏怎么假设，然后倒推出现金流和IRR。典型的投拓思维。","totalCalls":4,"totalTokens":820000,"sessionsAnalyzed":6,"topDomains":["地产项目投资测算","现金流模型与 IRR 分析","市场调研与竞品分析"],"roastTitle":"投拓测算界的扫地僧","projects":[],"gripHi":[],"gripLo":[],"quote":"这个 IRR 算出来 28% 不对劲，你检查一下土增税有没有漏算","oneLiner":"Excel 里有 47 个 sheet，VBA 宏的变量名叫 a1、a2、a3，但他的 IRR 从来没算错过。","activeDays30":0,"skills":[{"id":"3cd88fd0-689e-4b99-963c-0e5d09d5c185","title":"地产项目 IRR 测算框架","skillType":"1","callCount":4}]},{"name":"guanshoubing","slug":"Q0dctnnvbe","avatarId":10,"tagline":"guanshoubing是那种把系统状态看得很“硬”的人：失败要清、重试要净、开关要一刀切，界面状态要立刻跟上。和 AI 协作时，guanshoubing不爱铺陈，会用短句快速推进，但在关键决策点会给出清晰的原则边界。你能明显感到，guanshoubing对线上可控性和行为可追溯性有持续执念。","totalCalls":0,"totalTokens":699481,"sessionsAnalyzed":127,"topDomains":["支付与订阅后端治理","前端状态与交互一致性","任务重试与错误恢复机制","工程运维开关与风险控制"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"[$promptfolio-summarize](/Users/moyin/.codex/skills/promptfolio-summarize/SKILL.md)","oneLiner":"","activeDays30":0,"skills":[]},{"name":"尹孟琦","slug":"Q7E8aPo145","avatarId":8,"tagline":"尹孟琦像一个对结果很挑剔的项目负责人：先把角色和交付方式说清楚，再让 AI 去跑流程。产出一旦不过关，他会立刻追问底层模型并换方案。","totalCalls":0,"totalTokens":523634,"sessionsAnalyzed":6,"topDomains":["图像处理与抠图","多模型编排","README 驱动的项目生成"],"roastTitle":"","projects":[{"name":"图片处理助手"},{"name":"README 驱动的项目生成"}],"gripHi":["图片输出质量与背景去除","模型选择与接入路径"],"gripLo":["具体执行细节"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Robert","slug":"3VN03feeMK","avatarId":2,"tagline":"Robert是一位使用中文的Golang后端开发工程师。TA对代码质量有高标准，会主动让AI生成的代码经过审查。在问题处理上表现出色——能精准区分问题边界，不接受半成品，愿意推倒重来。","totalCalls":0,"totalTokens":500000,"sessionsAnalyzed":15,"topDomains":["后端开发","Golang","API开发","微服务","AI动画生成","ComfyUI"],"roastTitle":"","projects":[{"name":"动画自动生成系统"},{"name":"dataportal-fe"}],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"XQ QIN","slug":"yrU7h9FG8V","avatarId":10,"tagline":"XQ QIN 是那种会先把规则写进工具，再开始写第一行代码的人。TA 习惯从架构、规范和工作流的高度与 AI 对话，把模型当长期搭档，而不是一次性外包。哪怕在看似单一的组件需求上，XQ QIN也会自然地联想到性能曲线、权限模型和未来可扩展性。","totalCalls":0,"totalTokens":486408,"sessionsAnalyzed":1,"topDomains":["现代前端工程（React + TypeScript + Tailwind + Vite）","中后台管理系统与权限路由","高性能数据表格与大数据展示","AI 辅助开发工作流设计"],"roastTitle":"","projects":[],"gripHi":["工程规范与类型安全","架构与骨架设计","性能与可扩展性边界"],"gripLo":["样板代码与初稿编写"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"张子萌","slug":"wXBVPgM63u","avatarId":2,"tagline":"张子萌写需求的方式很像写规格：先定核心约束和不可修改项，再用子任务+验收要求把实现“钉”到一个可交付的形状。TA和 AI 协作时更像在教它建立边界，而不是让它自由发挥。","totalCalls":0,"totalTokens":481649,"sessionsAnalyzed":9,"topDomains":["工程落地/验收","移动/小游戏","产品机制设计","排行榜/社交"],"roastTitle":"","projects":[{"name":"项目-88ca0a"},{"name":"项目-710261"},{"name":"项目-b6408d"}],"gripHi":["规则/约束/验收口径"],"gripLo":["具体实现与代码细节","文档拆分/格式整理"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"kaven li","slug":"h6w0fuDVYO","avatarId":2,"tagline":"kaven li 像那种不爱空谈流程、但会默默把整条链路拧顺的人。他用 AI，不是为了陪自己 brainstorm，而是为了压缩实现时间、减少切换成本，并把现有知识库、组件库和发布流程都塞进同一套工作节奏里。","totalCalls":0,"totalTokens":461983,"sessionsAnalyzed":25,"topDomains":["React / TypeScript 前端开发","组件库工程化与文档体系","MCP 与 AI 编码工作流集成","认证登录与 Web 应用排障","移动端真机预览与跨端联调"],"roastTitle":"","projects":[{"name":"内部组件库与扩展工具链"},{"name":"AI 客服机器人示例"},{"name":"认证与登录实验项目"}],"gripHi":["任务边界与上下文控制","编码前知识来源"],"gripLo":["具体实现铺陈"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Nate Brooks","slug":"nate-brooks","avatarId":5,"tagline":"Epidemiological thinking: exposure, outcome, confounders. Starts with the research question, then figures out what data to pull.","totalCalls":0,"totalTokens":445000,"sessionsAnalyzed":5,"topDomains":["Epidemiological data analysis","R programming for public health","Data cleaning & wrangling"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the denominator? You can't show me a count without a denominator.","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Mike Chen","slug":"mike-chen","avatarId":6,"tagline":"Reader-first, task-oriented. Every page starts with 'what is the developer trying to do?' and works backward to the minimum content needed.","totalCalls":0,"totalTokens":372000,"sessionsAnalyzed":5,"topDomains":["API documentation","Developer guides & tutorials","Code example authoring"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"If the code example doesn't compile, the entire page is useless","oneLiner":"","activeDays30":0,"skills":[]},{"name":"linsanzhu","slug":"GahT3t0Cr5","avatarId":4,"tagline":"linsanzhu是一位在AI工具链中寻求工程化解决方案的开发者。TA不满足于使用现成工具，而是主动扩展工具边界，构建跨平台兼容的个人工作流。在与AI的协作中，TA展现出提示词工程师的特质——善于将模糊需求转化为精确指令，通过迭代优化达成目标。","totalCalls":0,"totalTokens":365000,"sessionsAnalyzed":265,"topDomains":["AI工具开发","提示词工程","数据处理","跨平台开发"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"BaoLei Li","slug":"iA3Ec4d_ao","avatarId":6,"tagline":"BaoLei Li 是那种会先弄清楚“数据去哪儿”“流程怎么跑”，再放心把自己交给工具的开发者。TA 一边试验新一代 AI IDE 和画像工具，一边不断确认本地分析、隐私边界和分享机制，用很工程师的方式审视自己的数字分身。","totalCalls":0,"totalTokens":364301,"sessionsAnalyzed":19,"topDomains":["AI 辅助编程与 IDE","开发者工具链与工作流设计","个人画像与协作网络","命令行与本地脚本集成"],"roastTitle":"","projects":[],"gripHi":["隐私边界","AI 工具链","工作流设计"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"柿子","slug":"OTAH-sphoy","avatarId":7,"tagline":"柿子 正在把 AI 代理当作可编程基础设施来构建。他通过 QQ 消息控制运行在服务器上的 AI，不是在用它，而是在给它装系统——每次验证完新能力，就写一条'记住'规则固化进 AI 的行为。他需求描述的方式透露了他的思维：第一句话就包含了边界情况和基础设施约束，几乎不需要后续修正。","totalCalls":0,"totalTokens":333437,"sessionsAnalyzed":7,"topDomains":["AI 代理编排与工具链","Python 自动化脚本","网络爬虫与数据采集","ML 基础设施","即时通讯机器人开发"],"roastTitle":"","projects":[],"gripHi":["AI Agent 编排","Python 自动化","网络爬虫"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Sherlock Holmes","slug":"Umd9QzTwy2","avatarId":7,"tagline":"Sherlock Holmes 做 Chrome 扩展与前端时，坚持「初学者能看懂」的写法：简单直白、关键行中文注释、无效无用代码必删。改东西前会先让 AI 说清主流程和核心逻辑，再问最简方案；布置复杂需求前会先问「你理解吗？不懂问我」。对死代码和过度设计习惯做减法，命名要语义化，状态默认保守再读持久化。","totalCalls":0,"totalTokens":333207,"sessionsAnalyzed":23,"topDomains":["Chrome 扩展开发","前端 Vue 组件与状态管理","网页翻译与内容脚本","录屏与整页截图","新标签页与小组件"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"刘文斌","slug":"HsBLBnZfNx","avatarId":8,"tagline":"刘文斌说话像产品经理兼工程师：先把交付物列清楚（界面、部署、文档），再补 IDE 一键启动；随后又两次跑全量画像总结，并尝试命令式登录——像是想把“做出来、跑起来、沉淀成画像”串成一条闭环。样本仍不多，但摩擦成本与可重复工作流已经很显眼。","totalCalls":0,"totalTokens":329422,"sessionsAnalyzed":1,"topDomains":["本地工具与脚本","Web 小应用交付","开发者体验与部署文档"],"roastTitle":"","projects":[],"gripHi":["交付范围与运行路径"],"gripLo":["具体实现与代码细节"],"quote":"","oneLiner":"","activeDays30":1,"skills":[]},{"name":"Rebecca Lin","slug":"rebecca-lin","avatarId":18,"tagline":"Content-first: start with the audience and the funnel stage, then work backward to the format.","totalCalls":0,"totalTokens":310000,"sessionsAnalyzed":5,"topDomains":["AI-assisted content creation","Email sequence copywriting","Brand voice prompt engineering"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"This sounds like AI wrote it. Where's the personality?","oneLiner":"","activeDays30":0,"skills":[]},{"name":"L1zHowo","slug":"W-StaLzUUX","avatarId":10,"tagline":"L1zHowo是典型的短指令、高执行密度协作型使用者：很少铺垫，直接给目标，然后快速看结果。TA对前端交付的要求不是停在可用，而是会追加一轮质量打磨，把任务从功能改动推向工程完善。和AI协作时，L1zHowo更像在推进迭代节拍，而不是做长篇讨论。","totalCalls":0,"totalTokens":303476,"sessionsAnalyzed":2,"topDomains":["前端表单交互","Vue3工程实现","TypeScript类型约束"],"roastTitle":"","projects":[],"gripHi":["交付质量与校验边界"],"gripLo":["实施路径与执行细节"],"quote":"","oneLiner":"","activeDays30":2,"skills":[]},{"name":"Maria Santos","slug":"maria-santos","avatarId":2,"tagline":"Audience-first. Starts with 'who is reading this and what do they need to decide?' then works backward to the data.","totalCalls":0,"totalTokens":289000,"sessionsAnalyzed":4,"topDomains":["Impact measurement & reporting","Python data wrangling (pandas)","Data visualization for non-technical audiences"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"What's the 'so what' of this chart? If a donor looks at it for 3 seconds, what do they take away?","oneLiner":"","activeDays30":0,"skills":[]},{"name":"qiang fu","slug":"ly-IhXcUJZ","avatarId":4,"tagline":"qiang fu是一位正在从API使用者向架构设计者过渡的AI工程师。TA用LangChain构建教育场景的对话系统，学习方式务实——要看到具体示例才能理解抽象概念，遇到不懂的直接问出来。工作方式是迭代式的，在实践中发现问题再回来补充需求，而不是追求一次性完美设计。","totalCalls":0,"totalTokens":283000,"sessionsAnalyzed":19,"topDomains":["AI对话系统开发","LangChain框架应用","向量数据库","教育科技","记忆系统架构"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"刘向阳","slug":"ygvjl_qE3R","avatarId":4,"tagline":"刘向阳是一位全栈工程师，专注于AI应用开发和现代前端架构，擅长LangChain、NestJS、TypeScript等技术栈，在MCP协议集成、流式输出、RAG系统等领域有深入实践。","totalCalls":0,"totalTokens":282558,"sessionsAnalyzed":167,"topDomains":["AI应用开发","全栈开发","RAG系统","微前端架构","CI/CD"],"roastTitle":"","projects":[{"name":"LangChain AI项目"},{"name":"NestJS集成项目"},{"name":"RAG电子书项目"}],"gripHi":["问题分析","类型安全","技术规范化"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Jianye Kou","slug":"EqDwsq23Su","avatarId":1,"tagline":"Jianye Kou是一个在企业安全分析领域深耕的技术人，擅长把复杂的业务场景（4A系统异常检测）转化为可执行的代码逻辑。工作中有一个鲜明特点：不是简单地下指令，而是先建立规范——让AI参考现有代码模式、统一命名规则、提供详细错误日志。技术选择上持实用主义，重构时坚持保留所有功能，对技术升级保持克制。全栈能力覆盖Python数据分析、Java后端和现代前端，但核心优势在于对安全领域的深度理解。","totalCalls":0,"totalTokens":278000,"sessionsAnalyzed":27,"topDomains":["企业安全分析","数据分析与挖掘","用户行为建模","全栈Web开发","数据库系统"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":1,"skills":[]},{"name":"李鑫晟","slug":"FYp34xRrJR","avatarId":8,"tagline":"李鑫晟��һλ�ѡ���ʾ�����̹��̡����������������ˣ�������趨�ھ��������ԡ��ṹ����ʽ����˽�߽磩�����ƶ�ϵͳ����ķ����۲����������������غ��еľ��߹켣��֤�������ԣ������ǵ��λش������Ƿ�˳�ڡ�","totalCalls":0,"totalTokens":245118,"sessionsAnalyzed":3,"topDomains":["��ʾ�����뼼����ϵ�","������˽���ȵ����ݷ�������","Cursor/VS Code ��̬��Э��������","��غϵ����빤�̻���֤"],"roastTitle":"","projects":[],"gripHi":["������Կھ� & �ṹ�ȶ�","��˽�߽�����ǰ��","֤��ȫ�ռ���ѹ������"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"嘉嘉","slug":"jiajia","avatarId":3,"tagline":"先想清楚这个事情的业务目的是什么、汇报对象是谁，然后倒推需要什么输出物。","totalCalls":0,"totalTokens":238000,"sessionsAnalyzed":4,"topDomains":["银行数字化转型项目管理","数据分析与Excel处理"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"这个报表明天早上开会要用，你帮我弄个能直接贴PPT的版本","oneLiner":"","activeDays30":0,"skills":[]},{"name":"李伟","slug":"LLAMJAvFPB","avatarId":10,"tagline":"李伟说话像在看分布式 trace：agent 多轮里每一跳发往模型的内容、何时 tool call、何时停，都要能对上源码与协议。装环境时又一秒切回终端民工，Homebrew 断线就换镜像重拉，但会坚持「告诉我发生了什么，命令我自己敲」——解释要给你，键盘不给你。","totalCalls":0,"totalTokens":213218,"sessionsAnalyzed":8,"topDomains":["AI Agent 与 LangChain/LangGraph 编排语义","OpenAI 兼容 API 与云上推理接入","Python 工具链与本地依赖/版本兼容","macOS 终端环境与网络代理排障","命令行工具链与包管理（Homebrew、Node/nvm）"],"roastTitle":"","projects":[{"name":"本地 Python Agent 学习实验"},{"name":"macOS 开发环境基建"}],"gripHi":["原理解释与可核对性","具体报错栈与环境归因","安装与命令执行"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"阿成","slug":"acheng","avatarId":22,"tagline":"从业务需求出发，先想「手工怎么算」，再想「怎么让Python帮我算」。","totalCalls":0,"totalTokens":205000,"sessionsAnalyzed":4,"topDomains":["工程量清单计算自动化","Excel数据处理（openpyxl/pandas）"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"你帮我写个脚本，把这个Excel里的工程量清单读出来","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Mia Suzuki","slug":"mia-suzuki","avatarId":20,"tagline":"Pattern recognition: look at the data across platforms, find what's working, understand why. Every insight needs to be actionable.","totalCalls":0,"totalTokens":198000,"sessionsAnalyzed":3,"topDomains":["Social media analytics & reporting","Sentiment analysis & NLP","Python API integrations"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"Engagement rate means nothing without context. Engagement rate compared to our category benchmark? Now we're talking","oneLiner":"","activeDays30":0,"skills":[]},{"name":"刘金磊","slug":"SYFjbjN410","avatarId":7,"tagline":"刘金磊是那种把调试当成系统工程的人：他不会满足于某个页面能显示，而是追问数据从哪来、成功语义是什么、以及线上/多端到底有没有同步。链路收敛之后，他又把体验推进到很细：从鉴权语义到按钮几何、从默认旅程到 TabBar 状态，都要符合可验证的标准。读完就会觉得，这个人做的是端到端一致性的“认真工程师”。","totalCalls":0,"totalTokens":184152,"sessionsAnalyzed":5,"topDomains":["跨端前端工程与数据一致性","API 契约与鉴权/错误语义","后端部署与发布边界控制","移动端交互与视觉规范","支付与订单链路可靠性"],"roastTitle":"","projects":[],"gripHi":["生产数据一致性与端到端同步","API成功语义（code + 201）","移动端视觉与交互规格验收"],"gripLo":["具体实现细节的落地"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"思思","slug":"sisi","avatarId":2,"tagline":"从专利权利要求出发。先搞清楚要分析什么技术领域，然后再想怎么获取数据。","totalCalls":0,"totalTokens":178000,"sessionsAnalyzed":3,"topDomains":["专利检索（关键词/分类号/引用）","专利数据爬取与清洗","Python基础脚本"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"爬虫又被反爬了？换个 User-Agent 试试","oneLiner":"","activeDays30":0,"skills":[]},{"name":"圆圆","slug":"yuanyuan","avatarId":14,"tagline":"先看法规要求，再看技术实现。每个数据处理流程都要先过一遍合规检查清单。","totalCalls":0,"totalTokens":167000,"sessionsAnalyzed":3,"topDomains":["医药数据合规","数据脱敏与匿名化","SQL审计查询"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"这个字段算不算个人信息？你先查一下个保法第四条","oneLiner":"","activeDays30":0,"skills":[]},{"name":"lvyiwn","slug":"rjU8XJAeTI","avatarId":6,"tagline":"lvyiwn是一个以“先跑起来”为原则的学习型全栈实践者：遇到问题先建立可复现的定位路径，再用最小改动恢复可运行状态。TA对沟通也有明确偏好（例如要求中文回答），希望把解决过程变成可以照做的步骤清单。","totalCalls":0,"totalTokens":156738,"sessionsAnalyzed":4,"topDomains":["全栈项目本地启动与联调","前端运行时排障（React/Next）","后端开发环境搭建（Node/Nest）","工程化学习与示例驱动实践"],"roastTitle":"","projects":[{"name":"Jira 类项目（本地启动）"},{"name":"百科类前端（React/Next）"},{"name":"百科类后端（NestJS）"}],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Monica Reeves","slug":"monica-reeves","avatarId":5,"tagline":"Top-down from the business question. Financial storytelling drives everything.","totalCalls":0,"totalTokens":142000,"sessionsAnalyzed":3,"topDomains":["Financial Planning & Analysis (FP&A)","Board Reporting & Investor Relations","Cash Flow Forecasting"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"Does this pass the smell test?","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Bennett Men","slug":"KOhcCqUgIr","avatarId":8,"tagline":"Bennett Men 在工程设计上的取舍很像“把边界写成契约”：先明确前后端职责与前端计算清单，再把性能/可靠性/安全当作默认规则写进文档。你在 Applet Engine 中也延续这种风格：用前后端集成测试启动约束、UIEvent 协议化（renderer/payload/locales）、以及 HTTP+WS 同步的兼容/时间语义来保证系统演进可控；同时用 gRPC/MQTT 明确通信职责边界，并在 ASL 开发规范里强制 plan 单一职责。整体上，你更像是把可靠性与长期演进成本提前写进架构选择，而不是上线后再修。","totalCalls":0,"totalTokens":139888,"sessionsAnalyzed":4,"topDomains":["前端工程架构与分层职责","实时数据与语音交互的工程化实现","浏览器端缓存、数据验证与版本迁移","部署安全默认值（CORS 与密钥合规）"],"roastTitle":"","projects":[],"gripHi":["前后端职责边界","缓存与数据治理生命周期","实时语音交互的非阻塞实现"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Andi Luo","slug":"andi-luo","avatarId":9,"tagline":"Workflow-first. Thinks in terms of 'what's the manual process and how do I eliminate it.'","totalCalls":0,"totalTokens":126000,"sessionsAnalyzed":2,"topDomains":["Podcast production automation","Audio processing with ffmpeg"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"Can you write me an ffmpeg command that does X?","oneLiner":"","activeDays30":0,"skills":[]},{"name":"lkj","slug":"SCoLidPHtn","avatarId":9,"tagline":"lkj偏好用中文沟通，并希望在执行前先把目标与步骤讲清楚。当前可用对话样本偏少，因此画像以“协作方式偏好”为主，暂不对技术栈与能力圈做强推断。","totalCalls":0,"totalTokens":120968,"sessionsAnalyzed":2,"topDomains":["AI协作流程","开发工具自动化"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":1,"skills":[]},{"name":"ShuangxuNian","slug":"dDj7ut9XGG","avatarId":3,"tagline":"ShuangxuNian 是一个偏向产品与工程交叉的开发者，喜欢先读文档再把任务拆成规范+UI+后端三段；善用 AI 作为审计/梳理工具，而不是直接抄答案。","totalCalls":0,"totalTokens":119003,"sessionsAnalyzed":4,"topDomains":["自然语言 UI 设计","前端交互布局","后端场景存储","AI 生成链路"],"roastTitle":"","projects":[],"gripHi":["前端交互和UI布局","AI 任务对齐"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"walking-malloc","slug":"75N2pJI7lp","avatarId":2,"tagline":"统计画像（本地统计）：建议 103,130 / 采纳 74,544（总体采纳率 72.28%）；拒绝率 27.72%；Composer 占比 67.22%；活跃 42 天。","totalCalls":0,"totalTokens":103130,"sessionsAnalyzed":42,"topDomains":["Go 后端架构","AI 协作开发","数据建模与交付"],"roastTitle":"","projects":[],"gripHi":["任务推进与执行节奏","结果一致性与质量"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"hshandyj","slug":"5FkI_7wpyA","avatarId":0,"tagline":"hshandyj 用 AI 打磨简历和项目表述时，坚持「表述要对得上代码」：技术描述与实现一致，不虚构不存在的实现，也不接受未实现的细节。会追问概念层次（例如 Gin 并发和 errgroup 的调用场景），要求选型理由真实、可辩护，并主动要求被面试官拷打，再反哺到文案与 Q&A。对连接池、KeepAlive 等实现细节会从「配置是否自洽」反推正确用法。","totalCalls":0,"totalTokens":98619,"sessionsAnalyzed":9,"topDomains":["后端服务与流处理","简历与面试准备","全栈与前端","数据与可观测性","并发与配置"],"roastTitle":"","projects":[{"name":"词云分词工程化服务"},{"name":"LLM 打标平台"}],"gripHi":[],"gripLo":["—"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"yg-striver","slug":"yM9Gv3bFmA","avatarId":9,"tagline":"是那种「先把需求写死再让 AI 动手」的人：函数的入参、返回值、边界情况得在提示词里讲清楚，不接受「实现 xxx 逻辑」这种空话。对 AI 生成的代码有固定的复查流程——排序逻辑、边界情况可以信，但并发和批量操作必须自己过一遍。Java 后端为主（Spring Boot、MyBatis、GORM），同时在做 AI 工程方向：LUI Agent、RAG、MCP 都有涉猎。会把 AI 当执行引擎用，接口自己定、架构自己把，具体实现交出去；发现问题先查自己的提示词，不是先怪 AI。对锁和并发有自己的判断，race condition 这种东西不靠 AI 发现，靠自己预判。","totalCalls":0,"totalTokens":98400,"sessionsAnalyzed":47,"topDomains":["AI 辅助软件开发","数据处理系统架构","性能优化与并发编程","提示工程与 AI 交互设计","代码质量与风险管理"],"roastTitle":"","projects":[],"gripHi":["AI 辅助开发","并发编程","性能优化"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"阿瑶","slug":"ayao","avatarId":22,"tagline":"视觉优先：先看效果对不对，再管代码写得好不好。","totalCalls":0,"totalTokens":95000,"sessionsAnalyzed":2,"topDomains":["UI 设计（Figma）","设计稿转前端代码"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"这个间距不对，设计稿是 16px 你这里是 14px","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Kit Tanaka","slug":"kit-tanaka","avatarId":15,"tagline":"Feature-first, fix-later. Jumps straight into building the cool part.","totalCalls":0,"totalTokens":87000,"sessionsAnalyzed":2,"topDomains":["Discord bots (discord.js)","Next.js web apps"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"wait it works?? let's goooo","oneLiner":"","activeDays30":0,"skills":[]},{"name":"enenkao1","slug":"x_E3Xv9Qiv","avatarId":1,"tagline":"enenkao1 是一个还在学习中、但已经有独立产品直觉的全栈开发者。他做的每个项目都有一两处非标准设计——URL密码访问控制、液态金属聚合逻辑、观光模式倒计时锁定——不是因为不知道标准方案，而是因为他先想到了自己的方案。他用 AI 的方式不寻常：不是让 AI 帮他写代码，而是让 AI 帮他理解，然后自己动手。","totalCalls":0,"totalTokens":85000,"sessionsAnalyzed":11,"topDomains":["前端交互与动效","全栈 Web 开发","桌面应用与打包","浏览器扩展开发","npm 组件设计"],"roastTitle":"","projects":[{"name":"个人技术博客"},{"name":"全栈电商平台"},{"name":"桌面 AI 文件管理助手"}],"gripHi":["前端开发","全栈开发","桌面应用"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Kaijun Jia","slug":"UkIjgSuJSK","avatarId":4,"tagline":"Kaijun Jia是那种把 AI 当合作者、但绝不会把“方向盘”交出去的人：环境、结构、原理这三件事他自己牢牢掌控，安装、执行、跑命令这种机械活才交给模型。他不爱改脚手架模板，更愿意倒查 TypeScript 版本和 VSCode 配置；写 demo 时一定要能在浏览器真跑一遍，再顺手让 AI给他补上事件循环、Hook 生命周期之类的长解释。跟他合作，你会发现他对“怎么想”和“改到哪里”为界限特别敏感——只要说清楚边界，他愿意让 AI 多干活；一旦触到原理或结构，他就会亲自下场把关。","totalCalls":0,"totalTokens":79676,"sessionsAnalyzed":22,"topDomains":["前端基础与浏览器行为","Vue3/React 应用脚手架与配置","AI Agent 工具链与本地环境","文档与系统提示词本地化","开发体验与工程化流程"],"roastTitle":"","projects":[{"name":"Vue3 学习与脚手架实验"},{"name":"前端基础与浏览器行为练习"},{"name":"AI Agent 工具链与提示词本地化"}],"gripHi":["开发环境与工具链配置","文档与提示词结构","原理讲解与知识建模"],"gripLo":["重复性命令与进程操作"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"ruiqi tang","slug":"ruiqi-tang","avatarId":10,"tagline":"从田间问题出发。先问'这个数据对种植决策有什么用'，再考虑怎么采集。","totalCalls":0,"totalTokens":68000,"sessionsAnalyzed":1,"topDomains":["农业物联网传感器集成","Python数据处理"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"数据一定要先存本地，网络断了再补传。田里的4G信号你不能信","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Roberto Firma","slug":"roberto-firma","avatarId":5,"tagline":"From first principles, always. Like a physics derivation.","totalCalls":0,"totalTokens":53000,"sessionsAnalyzed":1,"topDomains":["Condensed matter physics","Scientific data analysis"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"The plot needs proper axis labels with units. A graph without units is not a graph, it's modern art.","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Ethan Liu","slug":"4qiyp8-uqW","avatarId":3,"tagline":"Ethan Liu 是一位专注于仓储物流系统开发的开发者，擅长将业务规则转化为技术需求，主要使用 FastAPI 和 Vue 技术栈。","totalCalls":0,"totalTokens":50509,"sessionsAnalyzed":2,"topDomains":["仓储物流","成本优化","系统开发"],"roastTitle":"","projects":[{"name":"仓储装箱优化系统"}],"gripHi":["业务需求定义","技术选型"],"gripLo":["AI 协作"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Ggrryta","slug":"Iw1BA60tUu","avatarId":2,"tagline":"Ggrryta 是一位资深的分布式系统架构师，在微服务、实时数据流处理、API 网关、数据库分库分表等领域拥有深厚的积累。TA 强调架构的本质主义思维——不满足于表面方案，追求根本性的解决方案。TA 懂得根据业务特点做取舍，注重用户体验而非技术正确性。","totalCalls":0,"totalTokens":50000,"sessionsAnalyzed":4,"topDomains":["分布式系统架构","实时数据流处理","API 网关设计","数据库分库分表"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"li","slug":"1NERM8ICt6","avatarId":3,"tagline":"li 是一位能力维度极宽的前端开发工程师，从 C 端的高并发活动、硬核的 Cocos 小游戏交互，一路贯穿到复杂 ToB 配置化框架的搭建。你在代码里有一种强烈的『反脆弱』特质：一方面拒绝简单重复制造垃圾，用极强的抽象能力推行协议驱动；另一方面充满对异常突发状况的敬畏，把防爆、防泄漏、防降级刻在基因里。","totalCalls":0,"totalTokens":44767,"sessionsAnalyzed":1,"topDomains":["大型运营活动开发","企业级ToB管理系统架构","CocosH5互动与游戏"],"roastTitle":"","projects":[{"name":"企业级通用管理中台"},{"name":"Cocos H5 趣味互动矩阵"},{"name":"高并发运营活动引擎"}],"gripHi":["Cocos & H5性能优化","运营组件开发","ToB系统架构"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"lieflatmaster","slug":"2ONvCwkOn0","avatarId":4,"tagline":"lieflatmaster 是一个未雨绸缪的工程师，他把 AI 当作需要 SOP 管理的工具，而不是可以即兴发挥的艺术家。他会在动工前建好 checkpoint，精确到盘符级别规划目录，然后用\"先自检再联调\"的节奏来驾驭长任务。","totalCalls":0,"totalTokens":36268,"sessionsAnalyzed":4,"topDomains":["全栈开发","Java后端","Vue3前端","数据库设计","Docker容器化"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"博唐","slug":"GYabMZ2NEH","avatarId":10,"tagline":"博唐 treats AI tooling like infrastructure: get it installed, authenticated, and running, then ask for the full pipeline output. Their style is concise and execution-first, favoring commands that unblock progress over extended back-and-forth.","totalCalls":0,"totalTokens":35779,"sessionsAnalyzed":2,"topDomains":["AI tooling setup and automation","Developer workflow optimization"],"roastTitle":"","projects":[],"gripHi":["AI tooling setup and automation"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"asaad safi","slug":"vzwiiB7H_B","avatarId":10,"tagline":"asaad safi 是一名 AI 架构师，擅长构建 AI Agent 和计算机视觉系统。TA 的核心特质是'高内聚低耦合'的架构信仰和'先规划后编码'的工作方式——在写代码之前，必须先看到完整的计划书。TA 对异步编程有近乎偏执的要求，全链路必须 async/await，不容任何同步阻塞。","totalCalls":0,"totalTokens":29808,"sessionsAnalyzed":27,"topDomains":["AI Agent 系统","计算机视觉 (YOLO8)","边缘计算","智能客服"],"roastTitle":"","projects":[{"name":"美术班客服 Agent"},{"name":"塔吊横吊梁智能视觉检验系统"},{"name":"textClaub"}],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Larry Zhu","slug":"u6_NWuiS9N","avatarId":6,"tagline":"Larry Zhu是一位在 ProseMirror 深水区游泳的编辑器架构师。TA 用精确到行号的 bug 报告驯服 AI，用 before/after 示例定义交互规则，用「即时反馈 + 优雅降级」的哲学打磨一个 Markdown 即时渲染编辑器。把 AI 当执行层管理，自己把控方向、架构和审美。","totalCalls":0,"totalTokens":14508,"sessionsAnalyzed":17,"topDomains":["富文本编辑器架构","Electron 桌面应用","前端工程化","开源项目运营","技术写作与推广"],"roastTitle":"","projects":[{"name":"Milkup"},{"name":"AutoPilot"}],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"治涵孙","slug":"UfEEbdD1_H","avatarId":4,"tagline":"治涵孙 是一位拥有“架构师灵魂”的资深专利代理人。他以严密的量化基准和清晰的边界意识调教 AI，致力于将模糊的专业知识固化为精确的工程闭环。","totalCalls":0,"totalTokens":9906,"sessionsAnalyzed":3,"topDomains":["专利智能撰写 (Patent LLM)","提示词架构工程 (Prompt Engineering)","生物信息分析逻辑 (Bioinfo Reasoning)"],"roastTitle":"","projects":[{"name":"PatentAgent (专利智能撰写系统)"},{"name":"Biomni (生物信息交互系统)"}],"gripHi":["专利智能撰写 (Patent LLM)","提示词架构工程 (Prompt Engineering)"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"wubin","slug":"g6BM1dqDMu","avatarId":9,"tagline":"wubin 是一个以「透明」为标准理解代码的前端工程师——不接受框架的隐式决策，不满足于「能跑」，必须知道「为什么这样跑」。TA 写代码时脑子里始终跑着一个真实用户的模拟器，失败路径和成功路径在 TA 的设计里享有同等地位。高效率是前提，慢是 TA 会说出口的 bug。","totalCalls":0,"totalTokens":8500,"sessionsAnalyzed":99,"topDomains":["微信小程序开发","数据可视化前端","内容管理系统","React 组件设计","跨端框架工程化"],"roastTitle":"","projects":[{"name":"微信小程序（Taro + React）"},{"name":"CMS 内容管理系统前端"},{"name":"AI 数据洞察平台前端"}],"gripHi":["框架底层机制","UX错误恢复设计"],"gripLo":["后端API设计"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"LI LI","slug":"yKz5KX812l","avatarId":11,"tagline":"LI LI is just beginning to leave a trace in their local AI coding tools, with only a few short, command-style sessions so far, so there is not yet enough evidence from real problem-solving conversations to draw a strong portrait of their thinking. As LI LI starts using these tools for deeper design, debugging, and architecture work—where they explain trade-offs and push back on the AI—the portrait here will evolve into a much richer, more opinionated sketch of who they are.","totalCalls":0,"totalTokens":2673,"sessionsAnalyzed":4,"topDomains":[],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"yjsf216","slug":"_PtVzM2uEl","avatarId":7,"tagline":"yjsf216是一位专注于 Flutter 开发的工程师，注重代码质量和应用的可运行性，能够识别并解决复杂的代码问题。yjsf216关注版本控制和团队协作，注重应用的用户体验，要求导航功能准确、流畅。同时，yjsf216也注重项目的文档化和标准化，确保项目结构清晰可理解。","totalCalls":0,"totalTokens":1500,"sessionsAnalyzed":2,"topDomains":["Flutter 开发","移动应用开发","前端开发","项目管理","版本控制"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Time Tim","slug":"Pn_WFLGmQJ","avatarId":6,"tagline":"Time Tim 是那种会持续盯住结果的人：不满足于‘已经在做’，而要‘已经可交付’。当发现样本范围不够，TA 会立刻重设时间窗口，确保分析基础扎实。和 AI 协作时，TA 的特点是快速定目标、快速校参数、快速推进到最终产物。","totalCalls":0,"totalTokens":1141,"sessionsAnalyzed":2,"topDomains":["AI工作流编排","个人画像生成","数据范围校准"],"roastTitle":"","projects":[],"gripHi":["目标与交付验收"],"gripLo":["分析执行细节"],"quote":"","oneLiner":"","activeDays30":1,"skills":[]},{"name":"zwalker","slug":"24J8UlGizv","avatarId":0,"tagline":"zwalker是那种先写“验收标准”再让 AI 开工的人：目标清晰、约束明确、交付可检查。TA 对 UI 还原与响应式有明显执念，同时要求输出结构化、语气专业。遇到数据不足时会先补齐输入，再继续推进分析与发布。","totalCalls":0,"totalTokens":514,"sessionsAnalyzed":1,"topDomains":["前端开发与 UI 交付","设计稿还原与响应式适配","对话数据整理与分析流程","结构化沟通与交付标准"],"roastTitle":"","projects":[{"name":"设计稿驱动的前端页面生成"},{"name":"对话数据导出与画像生成流程"}],"gripHi":["UI 还原与响应式","结构化输出"],"gripLo":["开放式探索"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"zhangqin","slug":"bFcJNEc86p","avatarId":10,"tagline":"zhangqin更像是在做“可用且好用”的产品工程：TA 会把交互体验拆成具体验收点，要求行为稳定可预期。对 UI 视觉也有明确标准，宁愿花时间把品牌区做得符合主题，也不接受默认的将就。","totalCalls":0,"totalTokens":423,"sessionsAnalyzed":3,"topDomains":["Vue3 管理后台开发","前端路由与导航体验","聊天类交互（滚动/流式输出）","UI 品牌区与视觉一致性"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"wenliang wu","slug":"BsnSDdfPtH","avatarId":6,"tagline":"wenliang wu的提问集中在模块导出与导入方式的细节核对，说明TA更关心接口是否正确而不是泛泛描述问题。TA与AI的交互偏短指令风格，强调让工具快速对焦。样本较少，因此画像保持保守。","totalCalls":0,"totalTokens":275,"sessionsAnalyzed":3,"topDomains":["前端模块组织","路由配置与导入导出","IDE 内置 AI 协作"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"史伟明","slug":"GWY_pU3mGu","avatarId":4,"tagline":"No session data available. This is a placeholder profile.","totalCalls":0,"totalTokens":100,"sessionsAnalyzed":1,"topDomains":[],"roastTitle":"","projects":[],"gripHi":[],"gripLo":["General"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"gehao wang","slug":"eFT-NiIbFF","avatarId":3,"tagline":"gehao wang像是在把 AI 当成可塑的同事：你会先把问题的“本质与边界”讲清楚，再要求对齐到可落地的实现细节。你不满足于能跑的答案，更在意结构、可维护性与思路是否正确。","totalCalls":0,"totalTokens":37,"sessionsAnalyzed":2,"topDomains":["软件工程","前端/全栈开发","AI 协作开发"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"gao","slug":"Sfkug1BITv","avatarId":1,"tagline":"gao是那种会一边拆系统一边造词的人：看到AI输出趋同，他不抱怨，而是发明'框架句'来重新切割问题；看到模块没有真实价值，他不优化，直接判死刑。和他协作会感到一种持续的校准压力——他会不断把'做什么功能'的讨论推回'为什么要这样定义世界'。对'差不多'的容忍度趋近于零，但这种苛刻最终指向的是可迁移、可复用、能让别人也看见的思维方法。","totalCalls":0,"totalTokens":1,"sessionsAnalyzed":164,"topDomains":["AI产品方法论与价值架构","人机协作信息架构","认知框架抽取与建模","信息可视化与数据呈现","品牌叙事与产品文案"],"roastTitle":"The Invisible Architect","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"A product visionary using AI to cosplay as an engineering team — and somehow shipping faster than both.","activeDays30":0,"skills":[]},{"name":"lxh0113","slug":"5JSZpSXtla","avatarId":2,"tagline":"lxh0113 说话很“工程化”：先定边界，再定顺序，最后要可验证。TA 不满足于“看起来做了”，会追问文件有没有落地、构建过不过、线上状态为什么没变。","totalCalls":0,"totalTokens":0,"sessionsAnalyzed":2,"topDomains":["后端工程与可观测性","Java/Spring 生态集成","实时 Web（SSE）","开发者工具与身份画像产品"],"roastTitle":"","projects":[],"gripHi":["后端工程与集成"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"jiayan2713-ui","slug":"eqyVPLmtjh","avatarId":9,"tagline":"jiayan2713-ui 是一位注重技术架构和系统思维的开发者。在AI协作中表现出对Python项目结构和环境配置的严格控制，能够识别复杂问题的根本原因并将项目分解为清晰的模块化组件。\n\n偏好模块化设计方法，在实施前重视正确的环境设置。在技术决策中倾向于重构而非修补，展现出系统性思维和实际的项目经验。对机器学习工作流和API集成有实际理解，能够设计端到端的解决方案。","totalCalls":0,"totalTokens":0,"sessionsAnalyzed":28,"topDomains":["机器学习项目架构","Python开发","API集成","系统配置","数据科学工作流"],"roastTitle":"","projects":[{"name":"心血管疾病预测AI系统"},{"name":"模块化机器学习工作流"}],"gripHi":["技术架构与项目结构","环境配置与系统设置"],"gripLo":["代码实现细节"],"quote":"","oneLiner":"","activeDays30":0,"skills":[{"id":"11510487-e318-4562-b736-79870acd1cc1","title":"Prompt Engineering","skillType":"1","callCount":0},{"id":"8e75d354-f3ae-4b60-b180-952239fe2d3b","title":"AI-Augmented Product Design","skillType":"1","callCount":0}]},{"name":"niuzekun886","slug":"we99HwcU86","avatarId":8,"tagline":"# User Portrait\n\n## Core Characteristics\nThis developer had 697 deep conversations with AI over the past 30 days, showing:\n\n## Tech Stack & Domain\nMain active project: picture-book\nCore technologies: React, API, TypeScript, npm, 前端\n\n## Cognitive Style\nFrom 10 \"framework sentences\" (moments teaching AI how to think):\n\n- **Values taste**: Technical choices based not just on function, but elegance and clarity\n\n## Representative Teaching Moments\n\n1. Analyze the user's Claude Code conversation history from the last 30 days to build a portrait for promptfolio.\n\nSession files are in: ~/.claude/projects/**/*.jsonl (17 files total)\n\nYour task:\n1. Pars...\n\n2. [2026-03-12 17:07:13.591 +0800] INFO:\n    action: \"generateOutline\"\n    success: true\n    outlineId: \"4Za2IxKSVE14Oi14gADYk\"\n[2026-03-12 17:07:13.595 +0800] INFO:\n    method: \"POST\"\n    url: \"/generat...\n\n3. 你现在作为我的前端工程化助手，帮我在当前目录下创建一个基于 Vite 的 React+TypeScript 脚手架项目，严格遵循以下步骤输出：\n\n### 核心要求\n1. 项目名称：ts-react-promptfolio（和 promptfolio 评分适配，聚焦 TS+React）；\n2. 包管理器：npm（适配 Windows 环境，避免 yarn/pnpm 兼容性问题）；\n3. 技术栈：Re...\n\n4. 修改绘梦AI主页面：\n1. 标题改为「绘梦AI 绘出你的梦」（居中醒目）；\n2. 页面仅保留中间一个输入框，placeholder为：\n🔹 孩子的年龄\n🔹 想传递的主题（如勇气、友谊、接纳自己、环保……）\n🔹 喜欢的风格（温馨/幽默/诗意/冒险/拟人化动物等）\n3. 用户输入后默认提示文字永久消失；\n4. 移除其他所有元素，仅保留标题+输入框+生成按钮。\n\n输出完整页面代码。\n\n5. 在已完成功能基础上，只新增以下内容：\n\n1. 在「故事大纲」「故事详情」旁边加一个「缩略图」按钮。\n2. 点击「缩略图」可以查看已生成的图片。\n3. 点击图片可以全屏显示（铺满屏幕）。\n4. 点击「生成图片」按钮的流程：\n   - 先检查当前项目是否已经生成过图片提示词\n   - 如果有：直接用已有提示词调用图片接口生成高清大图\n   - 如果没有：先调用图片提示词agent生成提示词，再自动调用...\n\n","totalCalls":0,"totalTokens":0,"sessionsAnalyzed":17,"topDomains":[],"roastTitle":"","projects":[],"gripHi":["TypeScript","React","Full-stack"],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"alasong","slug":"MqgFNGhOKM","avatarId":1,"tagline":"alasong 是一位务实的系统构建者——在 AI Agent 和量化交易领域深耕，厌恶冗余，相信实证。TA 的'少说多做'不是沉默，而是用行动代替解释：快速尝试、快速中断、快速迭代。","totalCalls":0,"totalTokens":0,"sessionsAnalyzed":36,"topDomains":["AI Agent 系统","量化交易","工作流编排","数据工程","代码审查工具"],"roastTitle":"","projects":[],"gripHi":["代码复用","文档简洁性","架构选择"],"gripLo":["AI 执行"],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Yang Lu","slug":"Zcwsz15zRN","avatarId":4,"tagline":"Yang Lu 是一位追求极致效率的跨界创业者，同时经营实体餐厅和探索 AIGC 内容出海。TA 将 AI 视为可协作的「数字同事」而非工具，invest heavily in upfront protocol design——为 AI 定义完整身份、人格和工作边界。沟通风格 Concise Only，语言习惯中英混用，对术语精确性有执念。面对技术障碍时冷静务实，采用「先通后优」的迭代思路。","totalCalls":0,"totalTokens":0,"sessionsAnalyzed":12,"topDomains":["AI 自动化","内容创作","餐饮运营","跨境电商","技术架构"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"shao jie xu","slug":"oA626muMrG","avatarId":2,"tagline":"shao jie xu 是一位专注于 WEB 前端研发的开发者，深耕 Vue.js 技术栈。TA 对前端架构有深度理解，擅长组件化开发和前后端协作。TA 的话语简短但精准，平均仅36字。当AI偏离规范时，TA 会直接指出并引导参考正确示例。","totalCalls":0,"totalTokens":0,"sessionsAnalyzed":106,"topDomains":["Vue.js前端开发","组件化架构设计","前后端协作","前端工程化","UI/UX实现"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"Ricardo","slug":"H5UQfZ0R0L","avatarId":6,"tagline":"Ricardo 是一位实践驱动的全栈开发者，精通现代 Web 开发技术栈，擅长通过快速验证和深入理解工具特性来解决问题。在过去 30 天的 176 个 AI 辅助编程会话中，展现了从视频下载工具到低代码平台的多样化项目经验，以及对代码质量的执着追求。","totalCalls":0,"totalTokens":0,"sessionsAnalyzed":176,"topDomains":["全栈 Web 开发","Python 后端开发","Vue.js/React 前端开发","视频处理与下载","自动化脚本开发"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"shuyan0723","slug":"z8A6OVEy11","avatarId":10,"tagline":"shuyan0723 是一名注重用户体验的前端开发者，在实现功能时会从用户角度思考问题，而不是仅仅满足于功能可用。对技术选型有思考深度，重视代码的可读性和可维护性。","totalCalls":0,"totalTokens":0,"sessionsAnalyzed":8,"topDomains":["React/前端开发","路由守卫/认证","WebRTC","表单验证"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]},{"name":"qjh04","slug":"fB1Vh_TtTY","avatarId":11,"tagline":"qjh04是一位'先理解再动手'的全栈开发者。TA不急于让AI直接改代码，而是先让AI充分分析项目结构、技术栈和业务逻辑，这种克制的协作方式展现了资深开发者的节奏控制力。","totalCalls":0,"totalTokens":0,"sessionsAnalyzed":26,"topDomains":["Java后端开发","Vue前端开发","移动端Web应用","Go语言开发","全栈架构"],"roastTitle":"","projects":[],"gripHi":[],"gripLo":[],"quote":"","oneLiner":"","activeDays30":0,"skills":[]}]