OK-GR, Your Personal AI Race Engineer

• 

  Left side telemetry summary plots that greet user (driver) when they upload their race data.

• 

  Finished with the session? Download a coaching summary with steps to improve in the next session

• 

  You can ask your race engineer about your telemetry data and they will coach you on how you can improve.

Inspiration

• Most amateur GR-Cup teams don’t have a dedicated race engineer analysing their data or giving structured coaching. After researching typical team workflows, I realised how big this gap really is.

• OK-GR was born from the idea of giving every driver their own AI race engineer, a tool that makes drivers and teams more competitive.

What it does

OK-GR is an AI-powered race-engineering assistant that:

1. Automatically summarises race sessions, telemetry, and performance.
2. Visualises key metrics (speed–distance traces, G-G circles, traction envelopes, key lap metrics).
3. Lets you chat with your personal AI race engineer, who analyses your data and shows where time can be found.

In short: it brings pro-level driver coaching to amateur teams.

How we built it

• I built the UI using Streamlit, enabling fast iteration in Python.

• The AI core uses OpenAI’s ChatGPT, customised with a race-engineer persona and tool-calling for telemetry analysis.

• Supabase stores large telemetry files as Parquet for extremely fast loading and filtering.

• The entire stack; UI, data pipeline, AI agent, Supabase integration, and deployment, was engineered solo, from scratch and deployed on Streamlit Cloud.

Challenges we ran into

• Parsing messy real-world motorsport telemetry and normalising signals.
• Building robust summary functions and visualisations for large datasets.
• Integrating a customised OpenAI agent with real-time data processing.
• Handling multi-hundred-MB telemetry files within Streamlit Cloud memory limits.
• Debugging cloud deployment issues with large file loading and caching.

Accomplishments that we're proud of

• This is the first full software product I’ve ever built end-to-end, including GitHub, backend/frontend structure, async data pipelines, and cloud hosting.

• I built OK-GR in just a couple of days and learned an enormous amount doing it. I really enjoyed the process.

• Seeing the system analyse real race telemetry and generate actionable coaching feedback was a huge milestone and I hope to work on more motorsport projects in the future.

What we learned

• Efficient handling of large datasets with .parquet, caching, and column-level filtering.
• Designing an LLM-driven tool-calling system for domain-specific reasoning.
• UX considerations for engineering tools: speed, clarity, and responsiveness.
• File structuring and debugging.

What's next for OK GR

1. Integrate voice-to-speech coaching to allow hands-free interaction.
2. Share the tool with GR-Cup drivers for real-world testing and see how we can itterate on OK-GR.
3. I would like to also explore integrating OK-GR into consumer GR cars as a track-day training assistant.

Built With

• api
• openai
• python
• streamlit
• supabase