Start with the possibilities.
XGBoost estimates home, draw and away probabilities. A Dixon–Coles goals model adds likely scorelines and expected goals.
See the prediction APIIndependent project · Open source
Understand
the game.
See the reasoning.
Match predictions you can inspect. Football questions you can ask. An AI assistant that brings the data to the conversation.
Python + Kotlin. Built to run on your laptop.
THE GAME, EXPLAINED
One platform. Five leagues.
- Premier League
- Bundesliga
- La Liga
- Serie A
- Ligue 1
FROM A FIXTURE TO AN ANSWER
More than a prediction.
Follow the numbers, understand the factors, then ask the next question.
Know what moved the model.
SHAP breaks down the factors behind a prediction, with football-friendly labels. The API builds features from match history.
Read the explanation designA conversation with evidence.
The assistant retrieves reference material with citations and calls prediction, explanation, fixtures and season tools for football answers.
Explore the assistant toolsSEE THE WORKING PRODUCT
From API to
Android, in
three minutes.
Real backend responses, explanations, local AI answers and a native app that keeps saved data available offline.
Run it yourselfBUILT TO BE INSPECTED
The engineering is open, too.
26seasons
46,709 matches · 42 pre-match features
Serving refit: 2000/01–2025/26
- Test the future, not the training data.Season-based evaluation, walk-forward checks and published benchmark comparisons.
- Keep answers tied to their sources.Retrieval, tool outputs and abstention evaluation. Grounding is tested, not a promise of infallibility.
- Make the whole system reproducible.Versioned data pipelines, a FastAPI backend, Android tests and documented architecture decisions.
WORKING LOCALLY. BUILDING WHAT COMES NEXT.
A foundation for grounded football AI.
Built by Sumukha. The current assistant runs on Ollama; hosted inference and a private beta are planned. This site introduces the project—the interactive platform runs locally today.