Every one of these started as a question I actually had.
Each project sits where two or three of the things I love cross: soccer, jiu-jitsu, lifting, nutrition, AI and data. I take the question, build the thing that answers it, and use it. Three are live, and one more is on the way.
The question. How do you program strength work around mat time, and track both?
What I'm building. A program that combines jiu-jitsu and lifting, and maybe an AI layer that adapts the week's lifting to the mat time. Three of my passions in one place.
The question. People keep asking which supplements I take. Could I answer once, with the studies behind it?
What I'm building. A research library: each supplement I know and take, what the research says, dose and timing, and where the evidence is thin.
The question. How well could a simple, transparent model have called the 2026 World Cup?
What I built. An Elo-driven Monte Carlo simulator that replays the whole tournament 100,000 times and lines the runs up against what actually happened. Euro 2028 and Copa América 2028 run on the same code once their draws are known.
Split Decision
The question. Could I call UFC fights with the one family of models my studies skipped?
What I'll build. An XGBoost model run before every card: a win probability for each side of every bout. The MMA counterpart of IntSoccerPredictor, gradient-boosted trees in place of Elo and Monte Carlo, scored the same way against the real results.