2025 WDC: Lando Norris• McLaren

F1 Race Insights

A machine learning platform for Formula 1 race prediction, featuring 8 ML models, 10,000 Monte Carlo simulations, and SHAP explainability.

92%
Prediction Accuracy
Points Finish
8
ML Models
Ensemble Ready
0
Training Samples
2016-2025
0
Sim Features
F1-Level Detail
🎯 The Problem We Solve

F1 generates 1.5 billion data points per race weekend

Fans want to understand "who will win and why" but the raw data is inaccessible and overwhelming. Traditional media reduces this to pundit opinions, losing data-driven insight.

92%
Points finish accuracy (top 10)
85%
Top 5 finish prediction accuracy
98.7%
AUC-ROC for win classification

"As a new F1 fan, I want to understand what might happen before a race starts, so I can follow the action with context instead of confusion."

What We've Achieved

Engineering Excellence at Scale

Built from scratch with research-grade ML and production-quality engineering

98.7%
AUC-ROC Score
Win classification accuracy
8
ML Models
Including custom NBT-TLF
10K
Monte Carlo Sims
Per prediction request
68
Engineered Features
Driver, team, track, temporal

Custom Neural Architecture

NBT-TLF: Neural Bradley-Terry with temporal embeddings - not available in any ML library

Production MLOps

Model versioning, A/B testing, drift detection, and real-time monitoring

Fan-Facing Explainability

SHAP values exposed to users - F1 teams keep this internal

Beyond F1 Team Tools

Capabilities Real F1 Teams Don't Provide

Our platform offers research-grade features that go beyond what typical F1 analytics systems provide

SHAP Explainability

Real F1 teams keep model explanations internal. We expose SHAP values directly to fans.

🏆 FIRST-OF-ITS-KIND

10K Monte Carlo Sims

F1 teams typically run 100-500 simulations. We run 10,000 for every prediction.

🚀 20X MORE SIMULATIONS

Bayesian Uncertainty

Full probability distributions, not just point estimates. Research-grade uncertainty quantification.

📊 POSTERIOR SAMPLING

Counterfactual Analysis

"What if driver X had car Y?" - Causal reasoning unique to research systems.

🔬 CAUSAL INFERENCE

Technical Excellence

Built with production-grade ML engineering practices and modern web technologies

8 ML Models

From Elo baselines to Neural Ranking Networks (NBT-TLF) with temporal features

XGBoostCatBoostLightGBMRFNBT-TLF
View architectures

Monte Carlo Engine

10,000 simulation runs with temperature, tire degradation, safety cars, and DRS zones

TempTiresSCDRSWeather

SHAP Explainability

Interpretable predictions with waterfall charts showing feature contributions

WaterfallForceSummaryBeeswarm

Built With

Python
XGBoost, SHAP, PyTorch
FastAPI
REST API
Next.js 15
React 19
TypeScript
Type-safe
Tailwind
CSS
Docker
Containers
35+
Interactive Features
Engineering-grade tools
15K+
Lines of Code
Python + TypeScript
100%
Type Coverage
TypeScript + Python hints
CI/CD
GitHub Actions
Automated testing

Ready to explore F1 predictions?