Staff Data Scientist & Tech Lead at YouTube · Ph.D.
Staff Data Scientist and measurement methodologist. I design measurement systems and causal inference infrastructure for behavioral data at scale — connecting messy perception data to observable outcomes through tehcniques like Bayesian multilevel modeling, experimental and quasi-experimental design, psychometric measurement, and survey methodology.
Currently at YouTube, where I support measurement roadmap and implementation for Marketing and serve as a technical lead on the team. Previously at Google (Quantitative UX Research) and Meta (People Research), where I built measurement programs spanning advertiser trust, product experimentation, and the psychological impact of content exposure on human reviewers in trust & safety contexts.
Staff Data Scientist and Tech Lead for YouTube Marketing measurement. Drive the measurement roadmap and set methodological standards for how the organization quantifies brand impact on user behavior.
Design and validate survey instruments measuring latent perception constructs, linked to platform behavioral data. Build the end-to-end analytical pipeline from data extraction through psychometric validation (EFA/CFA) to causal impact estimation using experimental and quasi-experimental methods.
Developed a measurement framework for Advertiser Trust integrating attitudinal survey constructs with behavioral signals, adopted as a core metric across 12+ product areas. Quantified causal impact of trust perceptions on revenue-relevant outcomes using observational causal inference methods.
Built a first-of-its-kind measurement program for the psychological impact of graphic content exposure on human content reviewers in trust & safety. Developed novel survey instruments and behavioral indicators in partnership with clinical psychologists. Translated findings into global policy changes affecting thousands of reviewers.
Implementing Burt's composite measure of network diversity in R, with visualization using tidygraph and ggraph.
Replicating Stata's heteroskedasticity-robust standard errors in R, and why the defaults differ.
Mapping Brooklyn's bike lane network using OpenStreetMap data and R spatial tools.