Grokgoal
Machine-learning product
A football prediction platform built on its own model, with a public scoreboard that shows how accurate it really is.
01Context
Most football tips are confident guesses with no track record. Anyone can claim to be right; very few show when they were wrong.
02Approach
Make honesty the product: publish calibrated probabilities, measure every model against the market, and show the whole track record in public. A language model may explain a prediction, but it is never allowed to produce a number.
03Product
A web app in Danish and English with match predictions, a public model scoreboard, an 18+ gate and paid plans.
04System
A data pipeline that collects fixtures, results, odds, line-ups and weather, and keeps every raw answer before it is used. Ratings and a model that is trained, calibrated and tested, with a tripwire against future data leaking into training. A new model only goes live if it beats the one before.
Built with Python · FastAPI · Next.js · Redis · Stripe
05Outcome
The data foundation and the first calibrated model are done; the full model ensemble and the paid features are being finished.
Designed so every claim the product makes can be checked against its own public record.
06Services delivered
- Product strategy
- Data engineering
- Machine learning
- Web application
- Payments
- Project
- Independent Jinjr product
- Status
- In development