I design AI systems that fail safely — if the system can't verify its output, nothing ships.
What I've built:
- CodeReview AI — GitHub App that reviews PRs against your team's actual documentation. Every finding cites the source. Try it →
- TradeEngine — Autonomous trading agent with deterministic-first architecture. The math runs first, the AI runs second. (Just for me)
- Finance Assistant — 3-stage LLM pipeline with dual compliance gates for stock analysis.
Research:
- 22 provisional patents, 15 in AI safety and fail-closed architectures, 7 in physics
- 11 published papers, 8 on hallucination management, compliance pipelines, and verification systems, 3 on physics
The thread that connects everything: AI is probabilistic. My architectures make probabilistic components behave deterministically through verification gates, structured fallbacks, and fail-closed execution.