We plugged an LLM into raw health data. It just read the numbers back.

Something we noticed: LLMs are surprisingly bad at health data out of the box.

Give Claude or GPT a week of sleep and HRV data and you get "Your average HRV was 45ms, slightly lower than last week's 48ms." Cool, I can read a spreadsheet too.

The problem: raw LLMs don't reason about health. They don't get that declining HRV + poor sleep + high strain for 3 days straight is a pattern that means something.

So we built the Health AI Engine as an MCP server. Any LLM plugs in and gets actual reasoning tools: trend detection, anomaly flagging, cross-score patterns, personal baseline comparison.

Instead of "your HRV was 45" you get "strain exceeded capacity 3 of last 5 days, recovery down 23%, sleep consistency dropped since Wednesday, reduce intensity, estimated recovery to baseline: 2 days."

That's the difference between a chatbot and something actually useful.

We're live on Product Hunt. If you want to see more open-source health tools, an upvote goes a long way: https://www.producthunt.com/products/open-wearables