AI Agent Systems · Quantitative Research · Workflow Automation
If you find my work useful, you can support me via Afdian.
I build AI systems that run unattended in production — research agents, RAG pipelines, and quantitative trading infrastructure — and I publish the proof.
- rag-with-receipts — a RAG service where every answer must show its source, sentence by sentence, or get flagged. Runtime citation verification + an eval harness that ships in the repo. Runs keyless in 5 minutes; 15 tests, CI-green.
- finnews-extractor — Chinese financial news → structured events, with source grounding verified independently of the model: every extraction span and attribute value is re-located in the original text, or flagged as a hallucination. Golden fixture from a real LLM run; 16 tests, CI-green, keyless.
- quant-research-portfolio — methodology from production systematic trading: a live-vs-backtest divergence audit (including a risk overlay that silently locked 86.8% of out-of-sample paths), a four-threshold factor acceptance gate, and a negative-results log — plus a tested mini-framework implementing the gate.
- AI Agent Orchestration — multi-agent systems, LLM/RAG pipelines, MCP tool integration, confidence-gated automation
- Quantitative Systems — multi-factor models, backtesting engines, live trading pipelines (A-share & crypto)
- Workflow Automation — n8n / Dify based business automation, document extraction with source grounding
- Query-Web3 — product manager; AI-powered information service platform in the Polkadot ecosystem
- 15+ years across software engineering, product management, enterprise consulting (Infosys, Huawei projects), and quantitative finance
- MBA (ECUST) · MSc Economics & Quantitative Finance (SUFE)
Building autonomous research & trading infrastructure. Most of my active work lives in private repositories — the pinned projects above are the public, runnable evidence of how I build.