So I made a thing. With all the World Cup buzz right now, I got the itch to build something just for fun (honestly, I hadn’t done that in a while), so I made a little World Cup 2026 predictor. It looks like a cheesy 16-bit arcade game (pixel players, goal celebrations, the works), but there’s a real brain behind it.
I basically showed it every international match ever played, around 50,000 games going back over a hundred years, and it learned the patterns in who tends to beat who, and by how much. So when you pick two teams, it gives you the odds the way a bookie would: how likely each side is to win, lose or draw, plus a likely score.
The fun part: it can play out the whole tournament 5,000 times in a couple of seconds (basically rolling the dice over and over) and tell you how often each country actually ends up lifting the trophy. Or you can pick one team and ride with them game by game until they either get knocked out or win it all.
And it quietly updates itself twice a day, so it keeps learning as the real results roll in.
For the curious, the stack: Python for the model (LightGBM plus a Poisson-based goals model), plain HTML/CSS/JavaScript for the arcade front-end, Gradio and Hugging Face Spaces for hosting, and GitHub Actions retraining and redeploying it twice a day.
#WorldCup2026 #SideProject