Work
Systems, tools, and independent evaluations I have built and shipped. Each page covers the problem, the decisions, the result, and the parts that did not go smoothly.
Selected
- BatonDeveloper toolA kit that installs a reliable code-delegation workflow into Claude Code
- LoomaAI infrastructureLocal-first project memory for coding agents
- TabFM EvaluationModel evaluationIndependent evaluation that surfaced and fixed a multi-GPU crash in Google's tabular foundation model
- TurboQuant EvaluationSystems evaluationReproduced and debugged KV-cache compression on a 16 GB M1 Pro
- mddocsCollaboration toolGit-native collaborative Markdown, with an agent API
- World Cup 2026 PicksLive productSelf-hostable prediction pool with live leaderboards