I build LLM systems that are evaluated, not just deployed.

I design and build production LLM systems — retrieval, agentic orchestration, and document intelligence — with the same engineering discipline as any other production service: evaluated before every deploy, observed once running, built to fail predictably instead of silently.

Currently building the RAG Starter Kit in the open: an ingestion pipeline, a LangGraph retrieval graph, citation-grounded synthesis, and a DeepEval gate that blocks any deploy where retrieval precision regresses. Agentic orchestration, MCP, and document-intelligence systems follow the same pattern — architecture explained, not just shipped.