Projects
Things I've built, shipped, or started and occasionally finished.
-
asknum-eval
Evals for ask-a-question, get-a-number systems. A natural-language question goes in, a number from a governed data source comes out — this repo checks the number. Two tiers over one golden eval set: a tool tier that grades the query engine, and an agent tier that grades the full pipeline, including refusals, caveats, and relative dates.
open source
-
Corylus
The control layer for an automated agentic work pipeline: fast, reliable human judgment over everything the machines ship. Agents pick up tickets and ship work on a schedule, a judge agent reviews each attempt with a verdict and confidence score, and one screen puts the queue in front of you to approve, send back, or discard.
open source
-
Flightdeck
A mission-control dashboard for running multiple AI coding agents in parallel across many repos — one glanceable view of every session on your machine, grouped by project and agent, so nothing gets lost between terminal tabs. Dispatches new work directly, and picks up sessions started elsewhere without needing a wrapper.
unreleased, project still in development
