NFL Predictor

Ansh Malhotra
Published

0
  • Project creator

NFL Predictor is an ensemble prediction system that combines multiple data sources and analytical agents to forecast NFL game outcomes. The system is built on a FastAPI backend and React frontend, deployed on Railway.
It uses a multi-agent architecture with four specialized components:

Stats Agent — pulls historical and current team performance data from the ESPN API
Weather Impact Agent — factors in game-day conditions (wind, precipitation, temperature) with uncertainty quantification and abstention logic for low-confidence forecasts
News Sentiment Agent — applies NLP to analyze recent team news, injury reports, and media sentiment
Market Intelligence Agent — incorporates betting lines and Vegas spreads as a signal of collective expert wisdom

Predictions from all four agents are weighted and aggregated into a final ensemble forecast, achieving a reported ~74.7% accuracy across three weeks of live testing.