AI Agent Engineer · Shanghai / Hong Kong
I work across AI agents, RAG, LLM evaluation, and Web3 security—connecting research, engineering, and product judgment. This is where I share projects, technical notes, and evolving ideas.
02 / WORK
Practical work across generative AI, agent workflows, and security analysis.
Combining semantic AI analysis, deterministic program analysis, and historical attack knowledge to produce evidence-backed smart-contract risk verdicts.
A legal-memo workflow for lawyers that improves source retrieval and final memo usefulness through retrieval diagnostics, reranking, prompt orchestration, and expert evaluation.
Turning business requirements into compilable XML through data engineering, model fine-tuning, and multi-layer evaluation for reliable structured generation.
03 / NOTES
Writing on AI engineering, product building, and problem solving.
How Jarvis separates a traceable archive from active model context and compacts long conversations without deleting their history.
Read noteWhy asking for confirmation in chat is not an approval mechanism, and how Jarvis pauses, persists, and resumes tool calls before side effects occur.
Read noteWhy I built this site and what I plan to share here.
Read note