01 / React · Three.js
Birthday Tree Birthday Tree
沉浸式 3D 生日树体验，支持摄像头手势控制、照片上传与粒子树实时互动，适合展示创意交互和 WebGL 能力。 An immersive 3D birthday tree experience with camera-based gesture control, photo uploads, and real-time particle interaction.
AI 产品经理 · 求职中
AI Product Manager · Open to Work
陈熠
Chen Yi.
AI 产品经理
AI Product Manager
专注 AIGC 与 B 端 AI 产品，3 年经验。曾主导美团「美境 AI」B 端创意设计平台核心模块，完成从 0 到 1 的 AI 能力商业化落地。
3 years of experience in AIGC and B2B AI products. Led core modules of Meituan's "MeiGen AI" creative design platform, achieving full-stack AI commercialization from 0 to 1.
87%→4%
特殊菜品重生成率优化
Prompt Engineering 迭代
Prompt Engineering 迭代
Special dish re-gen rate
via Prompt iteration
via Prompt iteration
1,393/日
1,393/day
菜品生成日均调用量
接入拼好饭后
接入拼好饭后
Daily dish generation calls
after Pinhaofan integration
after Pinhaofan integration
93.6%
套餐图建券需求覆盖率
日均生图 1,487 次
日均生图 1,487 次
Combo image coverage rate
1,487 daily generations
1,487 daily generations
↓65%
营销图延展成本降低
十一战役 vs 纯人工
十一战役 vs 纯人工
Marketing cost reduction
Golden Week campaign
Golden Week campaign
核心优势Core Strengths
多维能力体系
Multi-Dimensional
Capabilities
从 AI 技术认知到商业化落地，构建完整的 AI 产品经理能力闭环。 From AI technical understanding to commercial deployment — a complete AI PM capability loop.
01
AI 技术能力AI Technical Capability
熟练 vibe coding，快速搭建前后端 MVP。熟悉 RAG、Multi-Agent 工作流、Prompt Engineering、多模态理解与生成。
Proficient in vibe coding for rapid MVP development. Familiar with RAG, Multi-Agent workflows, Prompt Engineering, and multimodal generation.
02
产品落地能力Product Delivery
从业务问题出发，完成需求洞察、PRD 设计、模型选型、Prompt 策略到灰度上线、A/B 验证的完整产品闭环。
From business problem to shipped product: user research, PRD, model selection, Prompt strategy, staged rollout, and A/B validation.
03
评测与迭代Evaluation & Iteration
擅长多模态生成评测，建立技术层 / 产品层 / 业务层三层指标，通过 badcase 归因持续迭代优化。
Expert in multimodal evaluation. Built 3-layer metrics (technical, product, business). Drives continuous improvement via badcase attribution.
04
跨团队协作Cross-team Collaboration
具备跨算法、工程、运营、合规多角色协同经验，能将复杂 AI 项目拆解为清晰的 SOP 与交付规范。
Cross-functional experience with algorithm, engineering, ops, and compliance. Converts complex AI projects into clear SOPs and delivery specs.
05
AI 商业化判断AI Commercial Judgment
具备 AI 产品商业化思维，关注降本、增效、转化业务指标，优先识别并规避技术导向的「伪需求」。
Strong commercialization mindset. Focuses on cost reduction, efficiency, and conversion metrics. Identifies and avoids tech-driven "fake requirements."
06
AI Coding 能力AI Coding
以 Claude Code + Codex 双 Agent 协同，将 PRD 直接转化为可交互前端界面，2–3 天完成端到端验证。
Claude Code + Codex dual-agent setup to convert PRDs into interactive HTML prototypes — end-to-end validation in 2–3 days.
工作经历Experience
AI 产品实践 AI Product Practice
从数据与 Prompt 工程到产品策略，覆盖 AIGC 产品完整价值链。 From data & prompt engineering to product strategy — covering the full AIGC value chain.
面向美团内部商家、运营、BD 等业务方的 B 端 AI 创意设计平台。作为应用层主 PM，独立负责三个核心模块（菜品生成、套餐生成、商品营销图）从 0 到 1 的完整生命周期管理。
B2B AI creative design platform for Meituan's internal merchants, operations, and BD teams. As lead PM, independently managed 3 core modules (dish generation, combo images, product marketing) from 0 to 1.
核心成果
Key Results
87%→4%
特殊菜品重生成率
Prompt 迭代优化后
Prompt 迭代优化后
Re-gen rate for special dishes after Prompt iteration
1,393/日
1,393/day
菜品生成日均调用量
Daily dish generation calls
93.6%
套餐图建券覆盖率
日均 1,487 次
日均 1,487 次
Combo coverage rate, 1,487 daily generations
↓65%
营销图延展成本降低
Marketing image cost reduction
+5~13%
商品曝光、点击、成交提升
Uplift in exposure, CTR & conversion
3天
3 Days
AI Coding 将对齐周期从 1–2 周压缩
Alignment cycle compressed from 1–2 weeks
核心工作
Core Work
- AI 工作流设计：主导设计「结构化参数 → Prompt 模板 → 多模型路由」三层工作流，将专业摄影语言抽象为 5 个业务方可理解的参数，设计三种交付形态覆盖 BD 建品、商家风格对齐、批量铺图三类场景。
- AI Workflow Design: Led 3-layer AI workflow design (structured params → Prompt templates → multi-model routing). Abstracted photography language into 5 business-friendly parameters with 3 delivery modes.
- 多模型评测：选取 200 个真实菜品 case，对即梦 5.0/4.5、Qwen-Image 等主流模型进行 3 维盲评，建立技术层 / 产品层 / 业务层三层评测体系。
- Multi-model Evaluation: Ran 200 real dish cases through 3-dimension blind scoring across major models (Jimeng 5.0/4.5, Qwen-Image, etc.). Built 3-layer evaluation framework.
- 多业务线对接：主导接入拼好饭、神枪手等业务线，形成「共建标准 + 双链路兜底 + 反馈闭环」的标准对接方法论并跨模块复用。
- Multi-line Integration: Led integration into Pinhaofan and Sharpshooter platforms. Established "co-build standards + dual-track safety + feedback loop" integration methodology.
- PRD 与 AI Coding：独立完成核心模块 PRD，使用 Claude Code + Codex 双 Agent 直接转化为可交互 HTML 界面，2–3 天完成端到端验证。
- PRD & AI Coding: Authored core module PRDs and used Claude Code + Codex dual-agent to convert them into interactive HTML prototypes — end-to-end validation in 2–3 days.
垂直于家居 / 室内设计行业的 AIGC 平台，目标是将设计师传统「效果图」流程压缩到小时级，覆盖室内设计、软装方案、产品营销图等场景。
AIGC platform for the furniture/interior design industry, aiming to compress the traditional rendering workflow to within hours. Covers interior design, soft furnishing proposals, and product marketing images.
核心工作
Core Work
- 提示词工程：建设 30+ 套结构化 Prompt 模板（风格定义 + 空间元素 + 材质组合 + 光影氛围），覆盖极简风、北欧风、现代风等主流装修风格；使用 MJ、SD、ComfyUI 批量产出风格化素材。
- Prompt Engineering: Built 30+ structured Prompt templates (style + spatial elements + materials + lighting), covering minimalist, Nordic, and modern styles. Used MJ, SD, ComfyUI for bulk stylized content generation.
- 多模态数据：主导家居图像与视频多模态数据描述体系建立，累计完成图像类数据标注质检 17,000+，大幅提升模型对家装风格的理解力。
- Multimodal Data: Led multimodal data framework for home images & videos. Completed 17,000+ image annotation and QA records, significantly improving model style understanding.
- 团队管理：管理培训 6 人标注团队，规划质检与标注分工，维持团队标注准确率 95% 以上。
- Team Management: Managed and trained a 6-person annotation team. Planned QA workflows and maintained annotation accuracy above 95%.
主要业绩
Achievements
17,000+
家装图像标注与质检条数
Home design image annotations & QA
30+
结构化 Prompt 模板套数
Structured Prompt templates built
95%+
团队标注准确率持续达标
Sustained team annotation accuracy
个人项目Projects
GitHub 项目 GitHub Projects
从交互体验、效率工具到 AI 小程序，把产品想法做成可运行的工程原型。 From immersive interaction and productivity tools to AI mini programs, turning product ideas into working prototypes.
02 / Next.js · Supabase
Flow List Flow List
现代化待办事项应用，覆盖拖拽排序、实时同步、优先级、图片附件、统计面板和用户数据隔离等完整任务管理能力。 A modern task manager with drag-and-drop ordering, realtime sync, priorities, image attachments, stats, and isolated user data.
03 / WeChat · Node.js · Coze
OOTD Mini Program OOTD Mini Program
基于微信小程序、Node.js 后端和 Coze Workflow 的智能穿搭推荐应用，按城市天气、性别和偏好生成每日 OOTD 建议。 A WeChat mini program powered by Node.js and Coze Workflow that generates daily outfit suggestions from weather and user preferences.
联系我Contact
期待与您深聊 Let's Connect
正在寻找 AI 产品经理的新机会。无论是产品策略探讨、技术方案交流，还是面试邀约，欢迎随时联系。 Actively seeking AI Product Manager opportunities. Whether for a product discussion, technical exchange, or interview — feel free to reach out anytime.