Inspiration
Race car drivers need help understanding their data. Professional teams have engineers who look at the numbers and tell drivers "you're braking too early" or "you need to pit in 3 laps." But most racers don't have this help - they just see confusing spreadsheets. We wanted to build a tool that acts like a race engineer, analyzing real race data and giving clear advice.
What it does
RaceIQ takes real race data from Toyota GR Cup races and turns it into useful information:
- Shows a 3D track with colors showing where you're fast (green) or slow (red)
- Calculates fuel usage based on your actual speed to tell you if you need to pit
- Analyzes your pace by comparing your lap times to other drivers
- Tracks tire wear by watching how your lap times get slower
- Finds where you're losing time by comparing your sectors to your best
- Compares performance across 7 different tracks to show your strengths
- Answers questions using Google Vertex AI - ask "where am I slow?" and get specific answers
Everything uses real data from actual races - 579 lap records with real times, speeds, and GPS coordinates.
How we built it
Data:
- Got CSV files from Toyota GR Cup races at 7 tracks
- 579 lap records with lap times, sector times, speeds, and GPS data
- Wrote Python code to read and clean the data
Backend (Python):
- Built analyzers that calculate fuel usage, pace, tire wear, and pit strategy
- Created a FastAPI server with 20+ endpoints
- All calculations use the real race data from CSV files
Frontend (React):
- Made a dashboard showing all the race information
- Used Three.js to draw the 3D track
- Added charts showing tire wear and performance
- Made it look good with animations
AI (Vertex AI):
- Created a knowledge base with 128 racing questions and answers
- Used Google Vertex AI to power a chatbot
- The AI can answer racing questions using the knowledge base
Tech used:
- Python, FastAPI, pandas for backend
- React, Three.js, Recharts for frontend
- Google Vertex AI for the chatbot
- CSV files for all the race data
Challenges we ran into
- GPS data was messy - had to clean it up to make accurate track maps
- Too much data - telemetry files are huge, had to use caching and sampling
- Started with fake data - rewrote everything to use 100% real race data
- Bad lap numbers - sometimes the system recorded lap 32768 (an error), had to filter these out
- Different file formats - each track's data was slightly different
- Slow 3D graphics - had to optimize the rendering to run smoothly
- Learning Vertex AI - figuring out how to structure the knowledge base and integrate the API
Accomplishments that we're proud of
- 100% real data - every number comes from actual race results, no fake data
- Works with 7 tracks - Barber, COTA, Indianapolis, Road America, Sebring, Sonoma, VIR
- Accurate calculations - fuel, pace, tire wear all based on real patterns
- Good 3D visualization - smooth animation with speed heatmaps
- AI integration - successfully connected Vertex AI with racing knowledge
- Clean code - well-organized, tested, and documented
What we learned
- Real race data is messy and needs lots of cleaning
- Caching makes everything much faster
- Racing strategy is complex - fuel, tires, pace all matter
- Building a good knowledge base is key for AI to give useful answers
- 3D graphics need optimization to run smoothly
- Full-stack development requires connecting many pieces together
What's next for RaceIQ
- Better AI - expand the knowledge base with more race data
- Deploy to Google Cloud - use Cloud Run, Cloud Storage, and Cloud Build
- Live race data - stream telemetry during actual races
- Weather integration - use weather data to adjust strategy
- Mobile app - make a version drivers can use in the car
- More racing series - support other types of racing beyond Toyota GR Cup
- Voice control - let drivers ask questions by talking
- Advanced features - traffic analysis, multi-race strategy, championship planning
Log in or sign up for Devpost to join the conversation.