Projects

evaltrack

A pytest plugin that gates and tracks your evals, and handles LLM flakiness without rerunning CI.

I think at least a subset of an agentic application’s evals should run and gate in CI, enough to know the main agentic behaviour still works. But LLMs are stochastic, and a single flaky eval means rerunning the whole pipeline.

evaltrack reruns only the evals that fail, tracks each case’s pass rate over time, and keeps every run in storage you own, so you can compare a branch to what is deployed. It works with pydantic-evals and DeepEval today.

Docs · GitHub

Clients

Currently: AI architecture advisory for a shipping company, and agentic AI development at a large energy company.