Gibbs AI

Inspiration

From the age of 5 I've been obsessed with motorsports, which lead me to race across the US for 25 years. I raced against some amazing natural drivers like Kyle Larson and Christopher Bell to see firsthand how critical the driver development process is to nurturing drivers to ensure they met their highest potential. I've also seen great talent go unseen and there are flaws that the process isn't more analytically driven. I wanted to build what I always wished existed: a data-driven system that could identify exactly where a driver was strong or weak, and provide actionable guidance for improvement. Having a product like Gibbs AI to continue to help motorsports use data similarly to baseball and golf to find the next superstars.

What it does

Gibbs AI is an AI-powered racing talent scouting and development platform that combines statistical rigor with personalized coaching:

Driver Rankings & Scouting: Global leaderboard powered by our validated 4-Factor Performance Model (R² = 0.895), explaining 89.5% of race outcomes. Breaks down our complex statistical rigor and make it easier for any driver / scout to understand.

Performance Dashboard: 4-page driver profiles with complete performance snapshots, race logs, skill breakdowns and driver development matching with action plans

AI Coaching Intelligence: Claude 4.5 Sonnet provides factor-specific coaching for Speed, Consistency, Racecraft, and Tire Management but real AI analysis

Comparative Insights: Learn exactly how to improve by comparing yourself to faster drivers in specific areas

How we built it

Data Foundation: 12 races, 34 drivers, 6,000+ laps with complete sector timing High-frequency telemetry (20-40Hz): speed, braking, throttle, G-forces, GPS Feature engineering pipeline: 12 performance metrics per driver/race Principal Component Analysis to extract 4 latent performance factors

Backend (Python/FastAPI on Heroku):

JSON file-based architecture for zero-latency API responses In-memory data caching for instant driver lookups Anthropic Claude SDK integration for AI coaching generation Docker containerization for reliable deployment

Frontend (React 19 on Netlify):

Toyota Gibbs Racing branding and design system Client-side routing with 5-page driver dashboards Framer Motion animations for smooth UX Responsive mobile-first design

AI Integration: Claude 4.5 Sonnet for coaching insights Factor-specific prompts with driver performance context Comparative analysis between drivers at the same track Practice plan generation based on weakness profiles

Challenges we ran into

We ran into two challenges which were around data size and finding our target audience. Dealing with telemetry data is difficult as there is a lot of it and you have to do research to understand. We decided it would slow down the experience if we focused on the data too much and then moved to aggregating it as we could. This allowed us to build the application to be fast and use the data in our AI insights. For the UX piece, there is a balance of whom would best use this and what level we would have to breakdown insights to get users to speak the common language of these insights. We were able to focus on the coaching elements rather then being overly technical with the analysis. Adding things like the sliders and ai summaries make the data digestable for drivers to figure out next steps and coaches with the quick insights to help drivers tailor training to become the best they can be.

Accomplishments that we're proud of

Statistical Validation: Built a performance model with R² = 0.895 and MAE = 1.78 positions, meaning we can predict race finishes within 2 positions on average, and explain nearly 90% of outcome variance.

AI That Actually Coaches: Rather than templated feedback, we integrated Claude 4.5 Sonnet to generate genuinely personalized coaching that considers driver-specific weaknesses, track characteristics, and comparative insights from faster drivers.

Full-Stack Production Deployment: Went beyond a prototype to deploy a real application with Netlify frontend and Heroku backend with Docker containerization.

Rewriting Racing Conventional Wisdom: Proved that we can make motorsports analytics easier to understand and fun. Having an easy navigation and video game like interaction should improve engagement with drivers on how to improve.

From 25 Years of Racing to Data: Transformed personal racing experience to build domain expertise into product and AI system desig.

What we learned

We learned that motorsports is moving in on sports like baseball and golf to find great talent with data. Getting to analyze this data and build a product around a field I have a lot of passion gave me a new appreciation of the sport I love. From a technical point I had to learn a lot around data architecture and sdlc process. Building a product isn't my background so I was invigorated to learn new skills and try to solve a problem that I've wanted to solve for two decades.

What's next for Gibbs AI

Would love to work more with Toyota to integrate this type of analysis with their other programs. They have significantly more data with more depth, but I feel like we've cracked a code where driver development could be improved an more transparent. All too often drivers spend time in the gym or on the sim and wonder how much that work will pay off. With Gibbs AI we help build focus and can predict how that will help them in their future results.