Local AI, agents and observability.

Self-hosted systems that run in production, not demos: RAG in regulated domains, LLM-in-the-loop agents, an MCP server a team queries daily. On owned hardware — private, zero cloud egress, $0 per query. Each one states what the design cost, not only what it does.

Part of my production work lives under confidentiality agreements. These are the projects I can open end to end — architecture, code, and the measurements behind them.