Pit Strategy Optimizer
Data-driven decision support for Formula 1 pit stop strategy
The Problem
Race engineers make million-dollar decisions in seconds with incomplete information
Stakes Per Race
A single pit stop error can cost championship points worth millions in prize money and sponsorship value.
Decision Window
Race engineers have seconds to decide: pit now or wait another lap. No time to model scenarios.
Underserved Users
Sim racers and F1 content creators need strategy tools but can't access team-level analysis.
User Pain Points
❌ No Validation
Teams can't easily validate post-race whether their pit strategy was optimal or if they left time on the table.
❌ Black Box Tools
AWS F1 Insights shows predictions but doesn't explain why. Engineers need reasoning, not just answers.
❌ Slow Manual Analysis
Analyzing lap times, tire degradation, and pit loss manually takes hours. Too slow for real-time decisions.
The Solution
Explainable AI that recommends optimal pit windows with confidence intervals
Tire Degradation Modeling
Linear regression models predict lap time degradation per track, compound, and fuel load. Interpretable and fast.
Pit Window Optimization
Simulates pitting on current lap and N future laps, applies pit loss and degradation, ranks by total projected time.
Explainable Recommendations
Explains why pit window opens, when degradation overtakes pit loss, and the cost of delaying or advancing the stop.
Historical Validation
Tests recommendations against actual F1 team decisions. Measures lap delta and alignment within ±3 laps.
Uncertainty Quantification
Sensitivity analysis shows how recommendations change with ±2s pit loss or ±0.02s/lap degradation uncertainty.
VSC Scenario Modeling
Models Virtual Safety Car scenarios where pit loss is reduced by ~50%, helping teams capitalize on race interruptions.
Product Development Process
How I built and validated this product
1. Problem Discovery
Identified gap in prosumer F1 analytics market. Teams have $10M internal tools. Fans have nothing actionable. Opportunity: accessible, validated strategy tool for 5M+ sim racers and content creators.
2. Scope Definition
Ruthlessly scoped to dry races and single-car strategy. Rejected multi-car game theory (exponential complexity), weather transitions (insufficient data), and real-time API (v2 feature). Focus = MVP validation.
3. Model Development
Built linear tire degradation models per track and compound using FastF1 historical data. Chose interpretability over deep learning complexity. Fitted models on 2023-2024 seasons with R² > 0.85.
4. Optimizer Implementation
Created pit window optimization engine that simulates N future pit scenarios, applies track-specific pit loss, and ranks by total projected race time. Outputs recommended lap with ±3 lap confidence interval.
5. Explainability Layer
Added rule-based explanation generation that describes why the pit window opens (degradation > pit loss), optimal timing, and trade-offs of waiting. Users need reasoning, not just lap numbers.
6. Historical Validation
Validated recommendations against actual F1 team decisions across 5 races (Bahrain, Monaco, Spain, Silverstone, Monza). Measured lap delta and alignment within ±3 laps to prove model accuracy.
7. Documentation & Testing
Wrote comprehensive PRD, assumptions document, and case studies. Built unit and integration test suite. Exported visualizations as PNG/HTML for portfolio presentation.
Success Metrics
How we measure product performance
Product Success Criteria
Validated against real F1 team decisions from the 2023-2024 seasons
✅ Recommendation Accuracy
Alignment within ±3 laps of actual team pit decisions
✅ Model Performance
R² > 0.85 for tire degradation predictions across all tracks
✅ Validation Coverage
Tested on 5 diverse circuits: street, permanent, high-speed, high-deg
Product Trade-offs
Key decisions and why I made them
| Decision | What I Chose | What I Rejected | Rationale |
|---|---|---|---|
| Model Complexity | Linear degradation | Deep learning | Interpretability > 2% accuracy gain. Users need to trust recommendations, which requires explainable models. |
| Race Conditions | Dry races only | All weather | 80% of races are dry. Wet/intermediate modeling adds months of complexity for 20% use case. |
| Strategy Scope | Single-car optimization | Multi-car game theory | Single-car covers 80% of value with 20% of effort. Multi-car interactions = exponential complexity. |
| User Interface | CLI + Python API | Web dashboard | Developer-first approach enables faster iteration. Web UI is v2 after proving core value. |
| Pit Loss Modeling | Track-specific constant | Dynamic traffic modeling | Insufficient data to model pit lane traffic accurately. Constant pit loss is good enough for MVP. |
Biggest Trade-off: Trust vs Accuracy
I chose a linear degradation model that achieved R² = 0.85-0.90 instead of a neural network that might have reached 0.92-0.95. Why? Because users (race engineers, strategists) need to understand why the model recommends a certain pit lap. A black-box model that's 3% more accurate but can't explain itself is useless in high-stakes racing decisions. This is the core PM lesson: the best solution isn't always the most technically sophisticated one.
Key Learnings
What I learned building this product
What Worked
- Starting with historical validation (proof before features)
- Documentation-first approach (PRD, assumptions, case studies)
- Ruthless scoping (dry races only saved 2 months)
- Building explainability layer (users care about "why")
- Comprehensive testing (unit + integration tests)
What I'd Change
- Talk to race engineers earlier (validated assumptions vs assumed)
- Build web UI sooner (CLI limits user testing audience)
- More visual design upfront (matplotlib plots aren't portfolio-ready)
- Document trade-offs in real-time (easy to forget reasoning later)
- Set up user feedback loop earlier (metrics aren't everything)
Biggest Lesson
PM work is 80% communication, 20% building. I spent too much time optimizing the degradation model and not enough explaining why it matters to users. The most technically impressive feature is worthless if users don't understand the value proposition.
Product Roadmap
Where this product goes next
Historical Validation
- ✓ Tire degradation modeling
- ✓ CLI interface
- ✓ Validation on 5 races
- ✓ Explainability layer
- ✓ Comprehensive docs + tests
Usability Improvements
- Web dashboard for non-technical users
- Multi-race comparison view
- Shareable race reports (PDF export)
- Interactive degradation visualizations
- User feedback collection system
Real-time Capability
- Live race API integration
- VSC/Safety car probability modeling
- Multi-driver team coordination
- Weather transition modeling
- Mobile app for trackside use
Monetization & Scale
- Freemium API model
- Integration with iRacing/sim platforms
- Content creator partnerships
- Team licensing (F2, F3, FE)
- Expand to endurance racing (WEC, IMSA)