At Epsilon, I built ClearSight, a multi-agent insights concierge that lets non-technical clients query campaign data in plain English. I added a verification gate that checks numeric claims against their sources and designed a 150-question benchmark to evaluate accuracy, hallucinations, and readability.
I also developed causal ML systems for creative effectiveness, including a creative-to-segment assignment policy projected to lift conversions by 7%. For synthetic control modeling, I built a scalable matching pipeline that paired 5M test IDs with comparable controls and achieved a 96% match rate.
