11 projects
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.
- Blue OmicsGenomics research platform for the UF blueberry breeding programJoined at 81 source files687, and the lead seat
- willitcallThe caniuse of local-model tool callingNobody had published the matrix32 rows, 3 servers
- PodiumVerified delegation for coding agents, where a shell command decides whether the work landedThe bot reports donethe runner decides
- LoomaLocal-first project memory for coding agentsRecall@3 0.621.00, on 0 third-party dependencies
- TabFM EvaluationIndependent evaluation that surfaced and fixed a multi-GPU crash in Google's tabular foundation modelCrashed on multi-GPUfix merged by Google Research
- TurboQuant EvaluationReproduced and debugged KV-cache compression on a 16 GB M1 Pro0% needle retrieval100% at 16K, 3.5x less KV memory
More
- BatonA kit that installs a reliable code-delegation workflow into Claude CodeBuilt the delegation kitused it to build itself
- mddocsGit-native collaborative Markdown, with an agent APIAn agent suggestsa person accepts or rejects, in git
- World Cup 2026 PicksSelf-hostable prediction pool for small groupsSkipped picks count as wrongno hiding in safe games
- ApplyScoreAI resume gap-analysis extensionPostings hide in Shadow DOM8 extractors, 1 fallback
- Football HubLive standings, fixtures, and scorers across 7 competitionsA rate-limited free tier7 competitions stay live