Hi, my name is
Windy Chen陈晓峰
AI Product Builder with technical depth.
3 career pivots · 4 verticals from 0 to 1 · Bridging tech and business
AI Product Builder · 12 years across 5 industries · Tech-Business translator
About
12 years across 5 industries and 3 career pivots: from technical translation (shipbuilding, chemical, naval) to big data and finance (Hua Xia Bank, industry research), to AI product management. Each pivot reached independent output within 3-6 months.
Business + interpretation background, self-taught in Python, TypeScript, Docker, RAG, MCP, LangGraph — the full AI stack. Not a traditional full-stack engineer, but an AI product builder who can rapidly understand new domains, build product infrastructure, and translate bidirectionally between technology and business.
There's a consistent pattern in what I've done: absorb a new domain fast, build the missing organizational infrastructure, and translate between two groups that can't directly talk to each other. From patent portfolios to data pipelines, from research frameworks to training programs — these things keep running after I leave.
Methodology
Most PMs talk about what they did. Methodology is how they thought.
Constraint-Driven Development
约束驱动开发
Use constraint docs, quality gates, and validation scripts to ensure architectural consistency in complex systems
In the RPS research production system: 54 constraint documents, 211 numbered rules, 90+ quality gates, 230 validation scripts — maintaining architectural consistency across 290 version iterations. Each issue is traced via RCA root cause analysis, fixed, and the constraint doc is updated to prevent recurrence.
Dual-Track for AI
双轨模型
Parallel exploration track (POC / technical validation) + delivery track (Sprint / milestones)
AI projects carry far more uncertainty than traditional software. In YingjiChemAI, I used 2-week cycles to POC technical feasibility first, then entered formal Sprint iteration only after validation. The exploration track continuously reduces uncertainty while the delivery track protects milestones.
Data-Driven Iteration
数据驱动迭代
Define KPIs → instrument → analyze → tune → validate in a closed loop
For YingjiChemAI, I defined three core KPIs at launch: query accuracy, response time, processing efficiency. User feedback data drove multiple iterations that raised query accuracy from initial 78% to 92%, with response time held under 3 seconds.
Research Intelligence Framework
研究情报框架
Taxonomy → Indicators → Sources → Matrix: a reusable 4-layer research methodology
Intelligence framework designed for Ghana country research: 11-dimension Taxonomy (Politics/People/Economy etc.) + 256 indicators (57 Key / 199 Supporting) + 6 IMF dataset mappings (WEO/IFS/GFS/BOP/FSI/DOT) + 50×6 cross-validation matrix. Swap in another country and the whole framework applies. This is methodology as asset.
Projects
8 AI products · 4 verticals · delivered from 0 to 1
YingjiChemAI
Full-stack AI research platform for chemistry
Chemistry research teams spent 2-4 hours per day searching literature and extracting key information
8-module AI platform: chemical extraction · 6-model Q&A · 3-way hybrid retrieval · GraphRAG · AI report generation · literature assistant · data center · application center
DocMind
2025.10 - presentIntelligent document parsing system
Qwen-VL based PDF-to-Markdown parser: 300 concurrency, multi-API-key load balancing, page-level checkpoint recovery, YAML chart metadata generation
MetaRAG
2025.10 - presentMulti-agent social science knowledge base
Declarative multi-agent orchestration (retrieve/summarize/reason/generate) with built-in Ghana research framework (11 dimensions / 256 indicators / 6 data sources)
PBIMCP
2025.10 - presentPower BI MCP Server
8-tool MCP Server supporting natural-language DAX queries, dataset management, chart generation, with OAuth 2.0 data security
AI Intelligence Analysis System
2023 - 2024Aerospace research institute
LLM+GraphRAG based intelligence system with multi-source auto-collection, aggregation analysis, and automated report generation. One of the earliest teams in China to introduce GraphRAG knowledge graphs
eZoo AI Tour Guide Glasses
2024 - 2025Operations analytics dashboard
Independently built a 5-dimension operations dashboard with TypeScript+Vite: device monitoring / visitor behavior / operations board / Q&A heat map / marketing tracking
E-commerce Big Data SaaS Platform
2020 - 2022Cross-border e-commerce customization
Translated business needs into technical requirements, coordinating with engineering to deliver data visualization, reporting, ERP analysis, blockchain traceability
Product Documentation
6 complete PRDs · six-layer structure · strategy/user/function/tech/metrics/management
YingjiChemAI PRD v3.0
DocMind PRD v2.0
MetaRAG PRD v2.0
PBIMCP PRD v2.0
AI Intel System PRD v2.0
eZoo Dashboard PRD v2.0
Click to view full PDF
Credentials
Intellectual Property
Certifications
Academic Honors
Contact
If you're hiring an AI PM, or just want to talk AI products