Football Power Rankings
Train a TensorFlow.js model on ESPN box scores and rank the NFL or FBS against FPI and the AP poll.
Football Power Rankings is a Next.js ranking engine that pulls every regular-season ESPN box score for a league and year, fits a small dense network on standardized point differential, and returns a power ranking you can read against ESPN. Pick NFL or FBS, pick a season back to 2000, and the server trains a 25-dimensional multilayer perceptron in TensorFlow.js — then lines our list up next to ESPN FPI (NFL) or the AP Top 25 (college).
The model is not a black box. Each team-game becomes a 25-D vector of yards, efficiency, returns, turnovers, and clock — everything except the score — then a 25→128→64→32→1 ReLU network predicts margin. NFL ranks by mean predicted differential; FBS adds a Simple Rating System so a blowout over a cupcake is not the same as a blowout over a contender. Live seasons shrink toward last year until enough games have been played. Try it at nfl-rankings-rose.vercel.app.