At NBCUniversal, I developed methods to measure hard-to-measure audience behaviors from imperfect sources and built an audience forecasting capability from the ground up. I benchmarked time-series models, engineered features, and scaled the pipeline across networks in a Databricks MLOps environment.
I also applied causal inference and experimental design to quantify effects and validate models, translating findings into decisions for cross-functional partners. I partnered with Product, Engineering, and Data Science teams to move models from proof of concept to scaled deployment, and co-invented a pending U.S. patent application for probabilistic audience deduplication and reach estimation.
At the University of California, Davis, I developed and published statistical methods for missing data and heterogeneous variability in mixed-effects models. My research used Bayesian inference, multiple imputation, and location-scale modeling, and I presented findings at national conferences and guest-lectured on Bayesian methods.
