Domify

At Domify I built two tools for the compliance platform. The first is a compliance playground that lets compliance officers test changes to their firm's criteria before they go live. It runs the updated criteria through Domify's annuities LangGraph pipeline and compares the results against the production baseline, both by structural diff and with an LLM acting as judge. It spans four repositories: a FastAPI service, a React interface, a gateway, and a read-only compute layer.

The second is a regression test suite. It lets you swap the model or prompt at any step in a multi-agent LangGraph workflow, then runs the full pipeline against a baseline to check what changed. When it finds a failure, it reruns the case to tell a real regression apart from normal model noise, and reports the cost, latency, and how much the output's wording drifted.