PPSFS AI - Sell Project

PPSFS AI — Sports betting analytics built on trading discipline, not tips.Most betting apps sell picks. PPSFS AI does the opposite: it applies the tools of a trading desk to sports betting, and shows its work.What makes it different1. It recommends by expected value, not by highest probability.Every bookmaker price is stripped of its margin (no-vig), compared against the model's own probability, and only the gap matters. A 70% favourite at a bad price is passed over; a 35% outcome at a generous price is flagged. That single rule is what separates value betting from tipping — and it's the core of the engine.2. All markets come from one simulation, so they can't contradict each other.A single Monte-Carlo run (10,000 simulated matches) produces the winner, over/under, both-teams-to-score, double chance and the most likely scorelines at once. Bookmakers price each market separately — that's precisely where exploitable inconsistencies appear. Verified: 1X2 sums to 1.0000, double chance to 2.0000, over/under strictly monotonic.3. Player props with real hit rates.NBA, NHL and Champions League. Per player: L5/L10/L20 hit rates, a weighted projection, the market line, and EV on both sides. Combined markets (points+rebounds+assists) are supported and settle on the summed box score. Under five games of history, no EV is shown at all — a probability drawn from three games isn't one.4. A public track record that can't flatter itself.Predictions are frozen before kickoff and graded against the real result. No look-ahead, no retroactive editing. Matches whose kickoff hasn't passed are structurally excluded from grading, so a bad data import can't inflate the record. Reconstructed predictions are labelled as such.5. Bankroll management, enforced.Fractional Kelly (0.25×), unit sizing, a hard 5% stake cap and a stop-loss. The app is built to stop a user from ruining themselves on a hot streak.6. A betting journal that fills itself.Sport → date → league → match → real pre-kickoff odds, per bookmaker. No team names typed, no odds typed. Bets logged as pending settle automatically once the match ends — P&L included. Parlays supported. Results displayed as a trading-style monthly calendar: green days, red days, weekly totals.Under the hoodBackend: Python / FastAPI, PostgreSQL, Redis, APSchedulerML: XGBoost with isotonic calibration, MLflow / Databricks model registry, River online learning, Optuna tuning, weekly automated retrainingFrontends: two React/TypeScript apps — a public app and an admin console — fully bilingual (EN/FR)Auth & user data: Supabase with row-level securityData: API-Sports (Pro), football-data.co.uk (25 years of closing odds), ESPN, official NHL/NBA APIs, SportsGameOdds for player propsAutomation: daily fixture, result, odds and props ingestion; every quota is enforced in Redis so no provider can be over-consumedCoverage: Premier League, La Liga, Serie A, Bundesliga, Ligue 1, Champions League, NBA, NHL.Honest status: live in free beta at betedgepro.ca, with a small user base. The public track record is real but the sample is still young — a few dozen graded matches since the season opened. Nothing here is inflated, and everything in the app is designed to make inflation impossible.