About The Bracket Lab

This started with a conversation. My friend Jack Gorman had an idea: what if you ran a bunch of different AI models against a March Madness bracket and actually tracked how they did?

I couldn't stop thinking about it. I work in AI consulting, and I'd never stress-tested these tools against something this clean — 63 games, public data, binary outcomes, a universal scoring system. March Madness is a near-perfect AI benchmark hiding in plain sight.

Standing up the site was the easy part. Claude Code and I got a working version live in under a week — eight models, a scoring engine, bracket visualizations, a leaderboard. The tool I used to build the thing is the same technology I'm testing. That felt right.

But "working" and "good" are different things. The real time went into the grind after that. Iterations on iterations — reformatting how brackets render, reworking data structures until they were clean enough to trust, documenting every decision so I could retrace my steps. This wasn't "make me a cool March Madness site." This was build, look at it, realize something's off, adjust, plan the next piece, sometimes take a step back, then iterate again until it felt right.

Mostly I built this because I wanted to see how far I could push AI as a prediction tool, and I'm hoping it helps my friends win their bracket pools. Not mine though — I'll fill mine out the old-fashioned way.