SMT Data Challenge 2026 Honorable Mention, along with Ethan Cappelleri, Evan Kurtz, and Tejas Rama. The prompt was to use SMT’s awesome player tracking dataset to build a fan experience that can be played on the in-stadium scoreboard. We chose to make a 3d animation library capable and a real-time catch probability optimized for being visually intuitive.
Resume bullets of what I did:
- Trained and optimized an XGBoost model to predict real-time catch probability every 50ms throughout outfield plays
- Linked model outputs to API for 3D catch probability cloud animations in ManimGL for on-demand video generation
- Constructed a dataset of 1300+ outfield plays with Polars, along with approximated walls, ball trajectory, and other features