Why I Built an AI System. (It's Not What You Think.)
I won my first fantasy championship in high school knowing nothing about football. My first question was why the players' pants didn't fall down when they got tackled. I still won. Fast forward to now — I work in tech, and I figured I could apply AI to the data nobody else looks at.
Why I Built an AI System. (It's Not What You Think.)
By GH | thegavfather.com
Artificial intelligence is going to change everything.
It's going to cure diseases. Solve climate change. Revolutionize education. End world hunger. Unlock the secrets of the universe.
I used it to win fantasy football.
I'm not apologizing.
How This Started
I won my first fantasy football championship in high school.
I knew absolutely nothing about football. I mean nothing. My first question when I started watching games was genuinely why the players' pants didn't fall down when they got tackled.
I still won.
I don't know exactly what that says about me. Probably something. But it planted a seed — if I could win knowing nothing, what could I do if I actually tried?
Fast forward to now. I work in tech. I spend my days thinking about AI, data, and how to find signal in noise. And at some point the obvious question hit me:
Why is nobody applying this to fantasy football?
Not the surface stuff. Not "here are last week's box scores with some analysis." The real data. The stuff buried in 22 years of NFL play-by-play that nobody has bothered to calculate because it requires actual computational work.
So I built it.
What I Built
22 years of NFL play-by-play data. Every snap from 2004 through the 2025 Super Bowl. 777,000 individual plays. 11,054 player-seasons.
175 features engineered from scratch — target share, air yards, EPA per play, referee penalty tendencies, coaching scheme fingerprints, dome vs outdoor splits, travel fatigue, game script adjustments, and about 160 other things that have never appeared on a fantasy cheat sheet.
Then machine learning models trained on all of it to answer one question: what actually predicts who scores fantasy points next year?
Not what sounds smart. Not what gets retweets. What the math says.
The answers were surprising. Some conventional wisdom held up. A lot of it didn't.
What the Machine Found
A few things worth knowing before your draft:
The single most predictive signal across 22 years of data — more powerful than target share, touchdowns, snap counts, or anything else — is what I call the Boom-Consistency Index. Players who score big regularly AND never completely disappear. The market rewards boom. The data rewards boom plus consistency. There's a meaningful difference and most drafters ignore it.
Third down targets — the stat everyone uses to find the quarterback's security blanket — actually predicts LOWER future production. The machine found this independently. I didn't expect it. The data doesn't care what I expected.
The bust prediction model is accurate 86% of the time. The biggest bust signal is players who scored a lot of touchdowns relative to their actual opportunity. Touchdowns aren't repeatable. The market prices players as if they are.
And 21 of 32 NFL teams changed offensive coordinators this offseason — the most coordinator turnover in NFL history — and virtually every public rankings site is still projecting 2026 performance based on 2025 scheme data that no longer applies. The Gavfather System accounts for every single change with position-specific adjustments based on each new coordinator's actual historical tendencies.
Why Fantasy Football
People ask me why I didn't use this for something more important.
Fair question.
Here's my answer: the same analytical frameworks that find edges in fantasy football find edges everywhere. Signal vs noise. What the market overvalues. Where conventional wisdom is quietly wrong.
I just happen to care about this particular problem. And I'm good at it. My pants-question rookie year championship says so.
Full rankings at thegavfather.com. Updated daily through camp and into the season. The model re-runs every time a player gets hurt, changes teams, or their coordinator gets replaced.
An offer your roster can't refuse.
— GH
The Gavfather System: 22 years of NFL play-by-play data. Machine learning signal discovery across 175 engineered features. Built by someone who still isn't entirely sure how the pants stay up.