At Norm AI, I was an early engineer on LEAP, a system that turns regulations into executable checks with auditable results. I built interfaces that let legal engineers inspect model decisions against source evidence and compare behavior across prompt versions.
I also built batch evaluation tooling that became the team's standard QA gate before prompt changes, cutting client onboarding time by 70%. By optimizing task infrastructure, I reduced production agent processing latency by 40% and compute cost by 50%; I’ve also created open-source tools for groundedness checking, LLM red teaming, evaluation, and streaming React interfaces.
