AI engineer with six years turning research into reliable products. Retrieval, evals, and LLM infrastructure built to survive real traffic. This site tracks everything I build, write, and wonder about.
Sub-50ms vector search over 40M documents. Hybrid dense + lexical ranking with a Rust serving layer.
Open-source framework for regression-testing LLM apps — golden sets, LLM-as-judge, CI gates.
Chat with your knowledge base. Grounded answers, citations, and confidence scoring on every reply.
Real-time defect detection on the factory floor, quantized to run on edge devices at 30fps.