I build models and data systems for professional sports. Most of my work is about injuries: how players and teams perform before, during, and after them.
I project how players, position groups, and teams perform around injuries in the NFL and NBA. I lean on Bayesian hierarchical models, which let players with little data borrow strength from the group and report a range of outcomes instead of a single guess. Every model is checked with walk-forward validation, so it is only scored on games it has not seen.
A health metric for players and teams, featured by ESPN and Fox Sports and used in analytics work with NFL teams. It turns messy injury reports and practice participation into a score that a coach, a fan, or an analyst can read at a glance.
Injury reports, practice participation, and player statistics arrive from many sources, mostly unclean and unstructured. I built the pipelines that clean and load them into SQL Server on a daily schedule, the AWS services that run them, and the REST APIs that serve the results to partners. Automated alerts email me when something fails, so problems surface fast.
At ScoreStream I worked on pulling real-time updates out of social media at scale, including text classification and spam detection across roughly 500,000 posts from more than 4 million users. I also built front-end widgets in TypeScript and React.
My UCLA thesis, in progress, builds a Bayesian model of how the economy relates to presidential election outcomes.