nathan@sf
role systems builder · researcher · student
focus CI/CD · distributed systems · agent evaluation
building noob-agent · better-ci · ci-cost · shadow-runner
Computational Sciences student at Minerva University building infrastructure for faster CI and more reliable AI-agent evaluation.
noob-agent — A reproducible benchmark for measuring whether AI agents can turn experience into executable Python skills that transfer to unseen game variations.
Python · SQLite · LLM evaluation · W&B Weave
BetterCI — A DAG-based CI engine with content-hashed caching, Git-aware job selection, and distributed workers. Skips up to 80% of unnecessary reruns.
Python · FastAPI · Redis · PostgreSQL
CI Shadow Runner — Reconstructs GitHub Actions dependency graphs and tests critical-path changes on live runners. Identified a dependency edge that skipped 186 test jobs and improved wall-clock time by 13% when removed.
Python · GitHub API · GitHub Actions
ci-cost — A vendor-neutral CI cost analyzer that audited 4,712 runner minutes across 200 GitHub Actions runs using eight versioned pricing models.
Python · GitHub REST API · CI/CD cost modeling