Race NeuroPilot

Inspiration

In the high-stakes world of the Toyota GR Cup, races are won by tenths of a second. While drivers fight on the tarmac, race engineers fight a battle of data—analyzing tire degradation, fuel loads, and pit windows in real-time.

We wanted to democratize this level of elite race engineering. Race NeuroPilot isn't just a dashboard; it's a "Digital Chief Strategist". We asked ourselves: What if we could give every amateur driver the power of a factory Toyota Gazoo Racing pit wall? We wanted to build an AI that doesn't just analyze, show charts, but actually speaks to you, making urgent strategy calls when the pressure is on.

What it does

Race NeuroPilot is a comprehensive, real-time AI race engineering platform:

1. ** AI Strategy Simulator:** Using Google Gemini 2.5, it runs Monte Carlo simulations (2,000+ futures) to predict optimal pit windows, tire compounds, and overtake probabilities based on live track conditions.
2. ** NEURO-SYNC (Wildcard Feature):** A sentient AI interface that speaks to the driver. Using Gemini 2.5 Flash Native Audio, it acts as a real-time Race Engineer, generating human-like, urgent voice commands ("Box now! Tires critical!") based on telemetry, rather than robotic text-to-speech.
3. ** Driver Coaching:** It analyzes telemetry to identify micro-sector gains, comparing driver inputs against physics models to offer specific advice (e.g., "Brake 5m later at Turn 4").
4. ** Automated Engineering Debriefs:** Instantly generates broadcast-quality post-race narratives and high-fidelity, dark-mode PDF engineering whitepapers for the team.

How we built it

We built a futuristic, "carbon-fiber" interface using React 19, TypeScript, and Tailwind CSS.

• The Brain: Google Gemini 2.5 Flash powers the logic. We don't use hardcoded if/else statements for strategy; we feed raw JSON telemetry to Gemini and let it "reason" through the physics and strategy.
• Data Ingestion: We engineered a custom Streaming CSV Parser to handle the massive 1GB+ TRD datasets directly in the browser. It reads files in 10MB chunks to prevent crashing, auto-detects schemas, and identifies the hero car.
• The Voice: For NEURO-SYNC, we used Gemini's Native Audio (PCM) capabilities. We capture the raw audio stream, process it through the Web Audio API, and visualize the frequencies in a custom HTML5 Canvas particle reactor.
• Visualization: We used Recharts for dynamic data plotting and Framer Motion for smooth, broadcast-style transitions.

Challenges we ran into

1. The 1GB CSV Problem: Loading the full TRD dataset froze the browser instantly. We had to implement a stream-based reader that parses data in chunks, filtering irrelevant rows on the fly to keep memory usage low.
2. Audio Latency & Quality: Standard browser TTS sounded too robotic for a high-end app. Switching to Gemini's Native Audio provided the quality we needed, but handling raw PCM streams in the browser required complex AudioContext buffer management to ensure smooth playback without "popping."
3. PDF "Dark Mode" Export: HTML-to-PDF tools struggle with modern CSS effects like backdrop-filter and gradients, often rendering blank pages. We solved this by building a dedicated "Sanitization Engine" that creates a simplified, vector-perfect clone of the DOM specifically for the PDF generation step.

Accomplishments that we're proud of

• NEURO-SYNC: We didn't just make a dashboard; we gave it a personality. Hearing the AI "speak" with the urgency of a real race engineer creates an emotional connection that static charts cannot match.
• The Aesthetic: The UI looks like professional software from 2030. The neon-on-black design language feels authentic to the TRD brand.
• Performance: Taking a massive CSV, parsing it, and getting an AI strategy recommendation in under 10 seconds entirely on the client-side.

What we learned

We learned how race engineers think—how small changes in braking points, throttle traces, and pit timing can completely reshape a race outcome. Building Race NeuroPilot showed us how to turn raw numbers into real-time strategy, actionable coaching, and meaningful insight.

What's next for Race NeuroPilot

• Live WebSocket Stream: Moving from CSV upload to a real-time data stream from the car's ECU.
• Vision Analysis: Using Gemini Pro Vision to analyze onboard camera footage for racing line deviations.
• Multi-Car Coordination: Linking multiple instances so a human crew chief can manage strategy for a 2-car team.

Built With

• gr-cup-dataset
• react