PitchPilot

Fyi: git refusing to merg, so if you'd like to play with the gaffa tool the functional version, email me: issakaamuli@gmail.com or tayobabsolugbemi@gmail.com :)

We built PitchPilot at ICHack to make football analytics genuinely useful in real time, not just a post match dashboard. Our whole approach was to treat tactical intelligence as a supervised learning problem: given what has happened so far in a match, can we predict the probabilities that matter and use them to value decisions? Using StatsBomb event data, we trained lightweight models to estimate pass completion, shot quality (xG), and win probability, learning these relationships directly from historical matches rather than hand coding rules.

At the action level, we compute xT (expected threat) by splitting the pitch into a grid and assigning each zone a danger value. Every pass or carry is scored by how much it increases threat: Delta xT = xT(end) - xT(start). That captures the value of ball progression beyond goals and assists. We then add risk reward by predicting the probability a pass completes and using that to estimate the expected value of attempting it, so ambitious passes are only rewarded when they are realistic.

At the match level, we build a compact information state each minute using time remaining, score difference, xG difference, and xT difference. Our win probability model learns P(win | state) from data and updates live as the game evolves, so you can see momentum shifts and whether a team’s performance is actually translating into winning chances.

To make it feel like an actual coach tool rather than just numbers, we introduced a “team gaffer” voice assistant. It speaks out key insights as the match changes, and it’s fully customisable using ElevenLabs so the voice and personality can match what a coach or team wants, whether that’s calm and analytical, high energy and motivational, or blunt and demanding. The gaffer can be tuned to the team’s style and preferences, so the same analytics can be delivered in a way that actually fits the environment.

We surface all of this through an interactive pitch visual (replays, heatmaps, passing networks) and clear, coach friendly insights that turn complex model outputs into actionable decisions. PitchPilot is our vision for a live tactical co pilot that helps coaches react faster and helps players improve with evidence based feedback.

Gaffer.mp4

Live Analytics

Performance Summary