Urdr: AI endurance coaching, hackathon winner

The problem

Endurance training data is rich but the coaching layer on top of it is usually generic. I wanted an app that ingests real training data and gives coaching that adapts to what the athlete actually did, and I wanted to prove I could build and polish it alone.

What I did

I built Urdr solo, end to end: training data ingestion, a local-first data layer synced with PowerSync, an AI coaching engine, and a web app on TanStack Start. The backend runs on Cloudflare Workers with queue-based processing pipelines written in Effect.ts.

The outcome

Urdr won Best Submission Using TanStack Start / DB / AI and placed 2nd overall at the PowerSync AI Hackathon 2026, judged on product polish, execution quality and how well sync, product design and the stack were integrated.