//How It Works
5 projects, 5 different answers to where the compute lives. Here's what's actually running behind each one.
ParlayGenIQ
LiveServer-side pipeline. Data lands on a schedule, models retrain in season, and every pick is graded in public after the fact.
Collect on a schedule
Automated pipelines pull schedules, lines, and box scores per league. 130+ cron jobs and 500+ scripts keep the warehouse current without anyone pressing a button.
Engineer features
Raw rows become predictive inputs — rolling averages, momentum, matchup and pace signals — built per sport rather than shared across all of them.
Train and retrain
XGBoost and Ridge ensembles, retrained during the season so the model tracks the current version of a league rather than last year's.
Rank, publish, grade
Model output is ranked into daily best bets, published before the games, then graded against the result. The record is the same one on the site — including the losing days.
Check the public track record — every pick is published before the games and graded after.
Two of these are archived or unfinished. They're here because how something was built doesn't stop being true when it stops running — and the retired ones sit in the graveyard with the reasons they were shut down.