AI tools, agent harnesses, and research pipelines. Most of it's open. The stuff that isn't is because I'm still figuring it out.
Provider-neutral AI agent harness. Swap between OpenAI, Anthropic, Ollama, or anything with an OpenAI-compatible API — the agent logic stays the same. Structured tool calling, JSON state, SQLite persistence. Built for agents that need to run reliably without a cloud provider.
VIEW ON GITHUBClosed-loop agent harness with structured outputs and step-by-step reasoning. Python-first. Define your agent's tools in JSON, wire them up, and get traceable, auditable agent runs. Good for agents that need to justify every decision.
VIEW ON GITHUBDaily automated research pipeline for Parkinson's disease. Every morning, a cron job pulls the latest PubMed papers, generates plain-language summaries in English and Spanish, and delivers them — automatically. No paywall. No algorithm. Just research, delivered.
VIEW ON GITHUBReal-time voice agent using WebRTC and a local LLM. No cloud API for the voice leg — the audio stays local. Built for agents that need to talk without latency or dependency on a third-party voice API.
VIEW ON GITHUB