What I'm up to

I'm wrapping up two years of shipping production voice AI: a multi-agent system that handles live insurance calls at sub-800 ms latency, plus the full-stack work around it, from the backend and APIs to the frontend and deployment.

These days I'm contributing to open source, most recently to freehire, an open-source job search engine, and building new things I can't stop thinking about. I just shipped 0.4.0 of claude-audit, a CLI that audits a codebase with fast static checks, plus an optional pass where Claude investigates the code with read-only tools. It's on npm and runs as a GitHub Action.

This month I also gave my Python packages a proper maintenance pass: code review, bug fixes, tests, CI, and token-free publishing. That shipped llm-ocr 1.2.0 (PDFs and images to Markdown with vision LLMs), async-limiter 1.2.0 (an asyncio rate limiter), and titans-unofficial 2.0.0 (a PyTorch implementation of the Titans paper).

Also building this very site. Designed to be machine-readable by AI models while still being a good experience for humans. Check /llms.txt.

What it takes to run AI agents on live traffic, not just in demos: latency budgets, evaluation, monitoring, and knowing when to hand off to a human. The gap between AI hype and what actually holds up in production is still enormous.

Interested in ideas that can genuinely impact people. The technology is the easy part. The hard part is finding problems that matter, and building tools that real users actually keep coming back to.

Developer experience as a design discipline. Good DX isn't just documentation. It's the API surface, error messages, sensible defaults, and respecting the developer's time.

End-to-end product pipelines. The journey from PRD to production: architecture decisions, deployment, observability, the boring bits that decide whether a product survives or not.

Evaluating voice agents.LLM-as-judge only gets you so far. I'm looking for better ways to measure whether a voice agent actually did its job on a real call.

AI tooling and the future of coding.The space moves weekly. I spend real time catching up on new models, agent patterns, and IDE workflows, so I don't fall behind the curve.

Professional standards and engineering practices. Code review discipline, testing strategy, documentation. The habits that separate side projects from real products you can hand to someone else.