WOW. Super impressed. My
@UFC gambling days are long behind me, but Iβve always wanted to build a Monte Carlo ML UFC fight predictor.
Every time I tried Iβd stall out on the same pain points β scraping UFCStats, cleaning and maintaining the data, engineering features, and then actually training and tuning the model was just way too much overhead for a βside project.β All of this was way too much for any AI coding agent to handle months just few months ago.
Fast forward to today and
@perplexity_ai Computer just oneβshot the entire thing for ~780 credits. It spun up a database of 4,455 fighters from UFCStats with full career stats and fight history, a Monte Carlo engine running 10,000 simulations per matchup with fatigue and finish probabilities, plus a logistic regression model scoring 15 features like striking differential, grappling, win rate, recent form, and physical attributes.
Then it blends both into a single combined probability, adds a confidence score, and on top of that gives you: winner win probability, method of victory breakdown, roundβbyβround finish odds, a clean stat comparison view, and even an EV calculator so you can plug in sportsbook lines and instantly see where the EV is.
A year ago this was a multiβweek build for a pretty advanced coder. Now itβs basically a prompt and some credits.
Feel free to try the app with the link below.