Tyche 2

AFL Prediction Engine

Tyche 2 is an AI-powered AFL prediction engine that uses Monte Carlo simulations, player-level modelling, and machine learning to generate probabilistic predictions for every match in the AFL season.

Named after the Greek goddess of fortune and chance, Tyche embraces the inherent uncertainty in sport. Rather than claiming to know the future, it models the probability space and gives you the most likely outcome along with a confidence measure.

Version 2 represents a complete rebuild — new data pipeline, new models, new interface. It processes 10,000 simulated outcomes per match, models individual player contributions, and continuously improves through post-match accuracy tracking.

How It Works

Monte Carlo Simulations

10,000 simulated season outcomes per prediction cycle, modelling scoring distributions and probabilistic match results.

Player-Level Modelling

Individual player predictions for disposals, goals, tackles, marks, and fantasy scores using historical form data.

Machine Learning Pipeline

LightGBM and XGBoost ensembles trained on 16,000+ historical AFL matches with 185 engineered features including Elo, rolling stats, venue, travel, and rest patterns.

Real-Time Data

Automated data pipeline pulling from AFL Tables, player stats, and injury lists to keep predictions up to date.

Ensemble Methods

Multiple model architectures combined — LightGBM, XGBoost, Elo ratings — with walk-forward validation across 2020-2024 seasons showing 67.4% accuracy.

Live Updates

Predictions recalculated as team changes are announced, with post-match accuracy tracking for continuous improvement.

Tech Stack

Frontend: Next.js 16 + shadcn/ui
Backend: FastAPI (Python)
ML Models: LightGBM + XGBoost + Elo
Simulations: NumPy Monte Carlo
Data: Parquet + Squiggle API
Hosting: Cloud Run (API + Web)

Built with ⚡ by Storm • Not financial advice • For entertainment purposes