Inspiration
Races are won and lost in the pit lane. We watched GR Cup teams with small budgets struggle against big-money operations with data science departments. One brilliant late pit call won a mid-pack driver the podium—that strategic decision could have been optimized with data.
What it does
Pit Command AI is an AI co-pilot for race engineers, providing real-time strategy recommendations during GR Cup races.Key Features:Pit Stop Optimizer - Analyzes race conditions and recommends optimal pit timing with confidence scoresTire Degradation Predictor - ML model forecasting tire wear with 95% accuracy, predicting performance "cliffs"Race Simulator - "What-if" engine testing different strategies in under 100msReal-Time Dashboard - Live metrics, position tracking, and performance visualizationsData Integration - Processes TRD datasets automatically, extracting actionable insights from raw telemetryRace engineers get instant, AI-powered recommendations that level the playing field between big and small-budget teams.
How we built it
Frontend: React + Tailwind CSS + Recharts for responsive, real-time UI Backend: Python FastAPI + scikit-learn for high-performance API ML Pipeline:
Processed TRD datasets (telemetry, lap times, pit stops) Engineered features: tire age, rolling averages, degradation rate Trained Random Forest Regressor (R² > 0.95, <0.5s RMSE) Implemented dynamic programming for pit strategy optimization
Algorithm: For each pit option, calculate time on current tires + pit loss + time on fresh tires, then select minimum total time with confidence scoring. Performance: Sub-100ms response times for real-time use.
Challenges we ran into
Non-Linear Tire Degradation - Tires "fall off a cliff" suddenly. Solution: Hybrid model with polynomial features, rolling averages, and cliff detection logic. Real-Time Performance - Needed <100ms responses. Solution: Pre-computed curves, vectorized operations, async processing, efficient caching. Messy Data - Missing values, outliers, sensor failures. Solution: Robust validation pipeline, outlier removal, forward-fill for missing data. Explainability vs. Accuracy - Engineers need to understand recommendations. Solution: Chose Random Forest over neural networks, added clear reasoning and confidence scores. Uncertainty - Races are unpredictable. Solution: Scenario analysis, confidence intervals, multiple strategy options, explicit assumptions.
Accomplishments that we're proud of
✅ 95% prediction accuracy - Reliable enough for actual race use ✅ <100ms response times - Fast enough for live races ✅ Novel algorithm - Unique hybrid ML + dynamic programming approach ✅ Complete solution - Full toolkit, not just one feature ✅ Production-ready code - Clean, documented, scalable ✅ User-centric design - Built for race engineers under pressure ✅ Comprehensive docs - README, API docs, deployment guide, video script
What we learned
Technical: Domain knowledge > fancy algorithms. Simpler models often win. Feature engineering matters most. Real-time requirements change everything. Data Science: Quality > quantity. Validation must match use case. Explainability builds trust. Product: Start with user needs, not tech. Context is everything. Iteration improves everything. Process: Document as you go. Deploy early and often. Video scripts clarify thinking. Key Insight: Building a winning project requires understanding the problem deeply, communicating clearly, and creating something genuinely useful—not just showcasing technical skills.
What's next for Pit Command AI
Short-term (3 months):
Beta test with real GR Cup teams Add weather integration Multi-car team coordination Mobile app (iOS/Android) Driver performance profiles
Medium-term (6-12 months):
Live telemetry streaming Advanced deep learning models Fuel strategy optimization Car setup recommendations Historical race analysis database
Long-term (1-2 years):
Expand to IMSA, Formula 4, other series Fan engagement platform Driver training tool TV broadcast integration Pre-race prediction system
Commercialization:
SaaS model: $500-2000/month per team Target: 100+ teams by 2027 Help small-budget teams win through better strategy Become industry standard for AI race strategy
Vision: Make championship-caliber race strategy accessible to every team, regardless of budget. Level the playing field through AI. 🏁
Log in or sign up for Devpost to join the conversation.