Quant researcher & agent engineer. HUST undergraduate building long-horizon research harnesses — persistent execution, environment feedback, auditable evaluation — and pointing them at markets and knowledge.
- Agent engineering — building a persistent research agent: typed tool loops, MCP, SQLite-backed state, resumable long-running jobs (browser-close-proof).
- Quant research — WorldQuant BRAIN consultant. Testing factor hypotheses and automating the idea → formula → backtest → diagnose → refine loop.
- On-chain research — reconstructing Polymarket fill-level ledgers from Polygon to audit reported P&L and study market-maker behavior.
- Personal knowledge system — a local retrieval stack (Flomo / Notion / past sessions → FAISS, BGE embeddings) that treats my own notes as a queryable database.
- First principles — most problems collapse once you drop to the layer where they're actually defined.
- Ergodicity — time averages and ensemble averages are different things; a lot of "long-term" advice quietly assumes the ensemble.
- Evidence over adjectives — don't tell me it's "robust"; show me the number and the denominator.
- Doing things that are useful, or interesting — ideally both.
- Syna — 72-hour hackathon demo: Rokid smart glasses → Android journaling app. Real hardware, installable build, honest about what wasn't connected.
Most of my current work lives in private research repos (agent harnesses, on-chain audits). Reach out if you want to see something specific.
Python · TypeScript · C++ · SQL · Kotlin
PyTorch · XLA/TPU · pandas · NumPy · BigQuery
Agent: AI SDK · MCP · FastMCP · SQLite · Vitest · persistent workers
Retrieval: FAISS · sentence-transformers
[email protected] — always up for talking markets, agent infra, or good books.