At Meta, I led the design and evaluation of an agentic ads-ranking experimentation platform, contributing to Meta-reported 2x average model accuracy across six models and 5x engineering output in its initial rollout.
I build rigorous offline-to-online evaluation, reproducible baselines, holdouts, uncertainty estimates, and human-approved promotion controls for large-scale ranking improvements across PyTorch, TorchRec, FBGEMM, and Apache Spark.
Previously at The Clorox Company, I built promotion measurement, product-review analytics, consumer lifetime value, and anomaly-detection platforms supporting marketing and consumer insights teams.
