What AI coaching
should feel like.
A coaching product athletes trust. Infrastructure clinicians can build on.
A coaching product athletes trust. Infrastructure clinicians can build on.
A coach that adapts as your life changes.
Goal, current fitness, anything you're working around.
Periodized program with phases, strength work, and weekly targets.
Traveling? Tweaked something? Tell the coach. Plan adjusts.
Running-focused licensed PT who already knows your history.
Your plan lives in the app. Zoom from the full program to this week to today's workout — the flow from strategy to action.
"I'm going skiing this weekend." Watch it adjust your plan around the trip — shifting workouts, dropping what doesn't fit, keeping you on track.
Tap any exercise, see the rationale tied to your injury and goals. Not generic advice — specific to you.
Mid-workout, it checks in. It will help you decide when it's safe to continue or time to back off. Real-time coaching, not a static checklist.
The coaching product works because the AI can't go off the rails. Constraints are enforced at the infrastructure level — not in the prompt.
Stages, fields, dependencies, temporal types. One YAML file, validated at startup.
Tools appear as prerequisites complete. The agent only sees what it should.
Dependencies, schemas, time windows — checked every call. Deterministic errors that help models self-correct — not 500s.
Every workflow is a YAML file — stages, fields, dependencies. At runtime, the server reads the YAML and generates MCP tools with Zod-validated schemas. Change the config, restart, new tools.
Tools unlock as the user progresses — later groups stay hidden until upstream work is done. Choose "training" and rehab tools don't just disable, they vanish. The agent only ever sees what's relevant.
Plans have date ranges. Weeks snap to ISO weeks. Daily workouts must land inside their weekly plan. The framework enforces the nesting — nothing exists outside its time window.
Ran product discovery alongside a co-founder. 60 user interviews, 500 people surveyed, and 20 demos with priced offering.
Interviews across athletes, running coaches, and physical therapists. Mapped injury fears, trust dynamics, and willingness to pay across personas.
Multiple Prolific surveys. Iterated toward the core segment: runners with race goals and injury history. Needed less than 1% of that segment to hit revenue targets.
Bundled PT eval, bring-your-own-eval subscription, and a full "pit crew" tier. Tested each with real-product demos and pricing conversations.
Defined success criteria before each batch of demos. By the end the learnings were prevalent, the data was unambiguous, and the decision framework was set before the results came in.
Each phase produced working software — not slides.
SMS-based onboarding prototypes, tested with PTs.
Inbox, video library, 3D skeleton overlays with joint angle tracking.
Shifted focus to athletes. Group intros, private intake, PT handoff.
Three packages tested with athletes in real demos and pricing conversations.
Full AI coaching app. 12 demos.
Original footage, joint angle overlay, full 3D skeleton — three views of the same exercise. Built for PTs reviewing patient form remotely.
Patient updates surface with AI suggestions. Tap a thread, see the draft — edit or send. Secure messaging that cuts response time without cutting the PT out of the loop.