I started building an FPL tool because I kept coming back to one question: what's the point of predicting the best player if signing him isn't actually the best move?
It gets complicated pretty quickly.
You have a budget, transfer limits, players you already own, upcoming fixtures, and the occasional terrible decision you made three gameweeks ago.
So I started building FPL Decision Engine.
The idea was to combine football data, AI reasoning, and rule-based checks to compare actual decisions rather than simply rank players.
One part I found particularly interesting was separating what AI suggests from what the system allows.
An AI model might recommend a great player. But if the transfer exceeds the budget or breaks the three-players-per-club rule, the recommendation should be rejected. Confidence doesn't make an invalid decision valid.
I also began exploring something else: what happens when my judgment disagrees with AI? Does it actually improve my decisions, or just make me more confident in them?
That research is still developing, and I haven't established that the engine improves FPL performance. There's plenty left to test.
The public repository contains a curated version of the code, architecture, and a worked example.
What started as a football project has made me much more interested in decision systems generally. Predictions are useful. Figuring out what to do with them is the harder part.