{"username":"junccch","header":{"name":"陈俊豪","tagline":"AI 产品经理 | Agent 产品实战 | AI Native Builder"},"personalInfo":{"email":"junshenqing23@163.com","phone":"(+86) 135-3550-7286"},"experience":[{"company":"美的集团","role":"AI 产品经理","startDate":"2025.11","endDate":"至今","bullets":["主导 3 款 Agent 产品从 0 到 1，负责产品形态定义、Agent Workflow、Skill 设计与跨系统工具调用；其中 2 款已进入生产环境。","能够直接设计并调优 Skills，使用 AI Coding 快速搭建可交互 Demo；基于模型能力与任务风险定义 Agent 执行边界，通过端到端评测与失败案例分析持续迭代产品。","围绕 HR 高频办公、AI Native 工作台与竞业审核等场景，将跨系统流程重构为 Agent 可执行任务链路，推动真实业务系统接入与上线评测。"],"tags":["Agent Workflow","Skill Design","AI Coding","RAG","Agent Evaluation"]}],"education":[{"school":"帝国理工学院","major":"公共卫生（统计学方向）","degree":"硕士","startDate":"","endDate":"2025"},{"school":"埃塞克斯大学","major":"经济学","degree":"学士","startDate":"2020.09","endDate":"2023.07"}],"projectsRecent":[],"projectsDetailed":[{"title":"OpenClaw 数字员工“小美”","type":"Agent 产品升级","startDate":"2025.11","endDate":"至今","bullets":["基于用户反馈与使用数据，判断继续优化问答准确率无法解决低频使用和任务断点问题；主导基于 OpenClaw 将“小美”从独立门户的问答助手，重构为接入高频办公入口、通过 Skills 调用真实系统完成任务的 HR 数字员工；上线后面向约 7500 名员工开放，DAU 5000+。","主导建立“面向完整用户任务的 Skill + 可组合底层工具”架构，将跨系统 API、RAG 与 CLI 封装为 Agent 可调用工具；直接设计并持续调优政策查询、考勤请假、人岗匹配等“问、查、办”Skills，使 Agent 能够主动追问、调用工具并完成任务闭环。","通过分析 Agent 执行链路，定位模型响应、冗余步骤及 Skill 发现瓶颈；推动模型切换、执行步骤精简与 Skill 发现机制优化，将政策查询首 Token 响应时间从 45-50 秒优化至约 12-15 秒。"],"tags":["OpenClaw","HR Agent","Skills","RAG","Performance Optimization"]},{"title":"AI Native 工作台","type":"AI Native 产品设计","startDate":"2025.11","endDate":"至今","bullets":["通过用户访谈与业务工作坊，从六大 HR 领域识别并拆解 30+ 适合 Skills；独立规划统一入口、领域隔离的 AI Native 工作台，推动接入 23 个业务系统与 5 个企业级数据库，预计年化节省成本 700 万元。","围绕招聘、人岗匹配、人才盘点等复杂场景，梳理业务规则、系统依赖与异常分支，将跨系统流程重构为 Agent 可执行的任务链路，支持 Agent 主动补充信息、调用工具并完成业务闭环。","针对查询、信息修改及业务审批等不同风险任务，分别设计 Agent 自动执行、用户确认与人工兜底机制；为招聘等跨系统任务定义任务完成率、工具调用正确率与结果准确性等验收标准，支持后续上线评测与迭代。","使用 AI Coding 独立搭建 AI Native 工作台可交互 Demo，以招聘全生命周期为代表场景，验证 Agent 跨系统完成端到端任务的产品形态；根据业务反馈迭代关键交互，推动方案通过评审并进入开发。"],"tags":["AI Native","Workflow Design","HR Systems","Prototype","Evaluation"]},{"title":"竞业审核 Agent","type":"智能审核 Workflow","startDate":"2025.11","endDate":"至今","bullets":["从 0 到 1 负责竞业审核 Agent，针对纯大模型审核的幻觉与一致性风险，采用“规则校验确定性问题、OCR 与多模态模型解析复杂材料、异常结果人工复核”的混合方案，输出包含审核结论、引用证据与风险提示的可复核报告。","基于历史案例构建评测集，通过案例回放与上线初期人工双轨复核，验证结果准确性及模型能力边界；将单次初审耗时从约 10-15 分钟缩短至 55-65 秒，月均节省约 500 小时人工审核时间。"],"tags":["Agent","OCR","Multimodal","Human-in-the-loop","Evaluation"]}],"skills":[{"name":"技能","items":["AI Coding","Prompt Engineering","RAG","Agent 评测"]},{"name":"语言","items":["英语流利","中文母语级","粤语母语级"]}],"contact":[{"label":"Email","url":"mailto:junshenqing23@163.com"},{"label":"Phone","url":"tel:+8613535507286"}],"meta":{"updatedAt":"2026-06-27T09:46:39.886Z"}}