Project overview
An adaptive AI running coach that syncs your runs and reshapes your training plan around real life.
Training plans assume your life is as consistent as a spreadsheet. Coach Ernie is a mobile app prototype that syncs your runs from Strava and builds a periodized plan on a rolling horizon, so a missed workout or a rough week reshapes what comes next instead of breaking the plan.
Under the hood it tracks training load with the acute:chronic workload ratio, a sports-science measure that flags when you're ramping up fast enough to risk injury. It also remembers facts about you as an athlete, so the advice gets more personal over time.
I designed the AI architecture around speed and cost. A fast, inexpensive model triages every message: simple requests like data lookups and workout swaps are handled immediately, while real coaching decisions are escalated to a more capable model. The result is that most requests come back faster and at a fraction of the cost.
The project started as a landing page with a waitlist and an experiment pulling data from Apple Health and intervals.icu, then grew into a React Native app. It's on pause as a working prototype.