How I Built an Agentic AI Coach That Turns Garmin Data Into a Training Partner

#devchallenge #hacktoberfest #opensource #agents

Hacktoberfest: Maintainer Spotlight

Intro:
I’ve been building an open-source project called Garmin AI Coach — a multi-agent endurance coach that analyzes Garmin health and activity data to generate adaptive training plans for triathletes, cyclists, and runners.

What makes it special:
It’s not just another fitness tracker dashboard. It’s an agentic AI system that actually interprets your physiological data (HRV, sleep, stress, load) and drafts a transparent, auditable plan you can inspect line-by-line. Every insight is reproducible — no black-box “trust us” recommendations.

How to contribute:

🧠 Help improve the agent reasoning or planning prompts.
🧩 Add new data connectors (Thryve, Garmin Health API, or Connect IQ).
🧪 Build demo datasets so new users can run the tool without real credentials.
🖼️ Create front-end visualizations for the generated HTML reports.

Issues labeled good first issue are beginner-friendly. I’d love to mentor new contributors, especially anyone interested in AI workflows, sports science, or wearable integrations.

Why I’m passionate:
As a triathlete and data scientist, I wanted a coach that explains why it suggests what it does. Garmin AI Coach is my attempt to merge transparency, personalization, and open science.

Links:
🔗 GitHub — leonzzz435/garmin-ai-coach
📰 Medium story — “I Fired My Garmin Coach and Built an AI to Train for an Ironman 70.3”

Top comments (7)

Yeahia Sarker • Nov 5

This is awesome, Leon.
really love how you made the agent workflow transparent “Nerds and Coaches” is such a clever design.
We’ve been exploring similar ideas around multi-agent reliability while building GraphBit
, and your approach to reproducibility really stands out.

Would love to chat sometime about how you’re managing agent state and coordination — super inspiring stuff!