Turning Strava telemetry into a live sports broadcast

My recent 10K data showed heart rates hitting 197 bpm, but looking at a flat graph doesn't capture the actual effort of a race. I wanted a way to turn those raw .gpx and .tcx telemetry files into something that actually felt like the event, so I built Race Recap Radio.

The pipeline works like this:

1. The app parses the file telemetry locally in your browser.
2. A summarized version of the intensity data is sent to Gemini to write a play-by-play script.
3. ElevenLabs converts that script into a high-energy sports announcer voiceover.

Instead of just seeing a spike in my splits, the AI narrates the surge as it happens. It makes the data feel like a cinematic broadcast rather than just a collection of numbers.

I prioritized data privacy by keeping the heavy lifting on the client side. Your full GPS traces stay in the browser; only the necessary telemetry summaries reach the LLM for script generation.

You can test how the Gemini/ElevenLabs combo handles high-intensity data here:

https://race-recap-radio.vercel.app/

I’m interested to see if Claude could manage a more natural narrative flow, but for this specific announcer persona, the current stack is hitting the mark. If you have a demo or a specific model setup for parsing telemetry, let me know.