Work / Alpine Velo

An AI coach that shows its work.

Alpine Velo is an iOS app I designed and built end to end. An agent reads every ride, plans the training, and explains its reasoning. A second model grades its answers in production, and it cannot change a rider's plan without approval. Every RF Digital principle, running live in a shipped product.

Three iPhones running Alpine Velo, showing insights, the day's training plan, and coach chat, beside a Download on the App Store badge.
The product

A coach in your pocket, not a calendar on your wall

Most training apps hand you a rigid plan and count how well you complied with it. Miss a week and you're "behind". Alpine Velo works the other way around: the coach looks at what you actually rode, adapts the plan to your real life, and tells you why.

Under the hood, the coach is an AI agent with real capabilities. It can query your full ride history, compute what your legs can do, and update your plan. But it's never unsupervised, and it never changes your data without your say-so.

  • Live on the App Store, with riders using it daily
  • Rides sync from Strava, Wahoo, and Hammerhead automatically
  • Every AI answer is quality-checked by a second AI
  • The coach asks before it changes anything you own
In the app

What riders see

Alpine Velo home screen showing a coach-written daily briefing and the next planned ride.
The day's plan, written by the coach
Chat screen where the AI coach answers a training question using the rider's own ride history.
Coach chat with real data behind it
Ride analysis screen with power and heart-rate charts and zone breakdowns.
Every ride analyzed automatically
Weekly training plan screen with editable coach-generated sessions.
A training week you can edit
Insights screen showing fitness, fatigue, and form trending over time.
Fitness trends in plain terms
Behind the app

The half of the build nobody demos

Shipping the product is half the job. Knowing whether it works is the other half. So there is a second application behind Alpine Velo that riders never see: an internal dashboard for watching what the AI actually does in production.

Every coach answer is stamped with the model and the prompt version that produced it, and can be rated in a review queue. When I change a prompt, I judge it against real conversations instead of a hunch. Cost per call sits on the same screen as engagement, so quality and spend get read together.

  • Every AI answer stamped with the model and prompt version behind it
  • A review queue for rating real answers, so prompt changes get judged on evidence
  • A bug radar surfacing anomalies across all users, not one report at a time
  • API cost tracked next to engagement, on the same screen
  • Onboarding funnel with drop-off at every step

This is the same thing I build for clients. A system you cannot observe is a system you are trusting on faith.

Why it's here

What this proves about working with RF Digital

SITE

Design and build, end to end

The whole product is designed and built from scratch: the app, its design system, the marketing site, and the internal dashboard I run it from. Every screen, every state, every word.

When I say I'll design and ship the whole thing, this is what that looks like finished.

FLOW

Automation that runs itself

Rides arrive from Strava, Wahoo, and Hammerhead on their own. Analysis runs the moment a ride lands. The morning briefing writes itself overnight. Nobody presses a button.

The same pattern as client work: multi-step processes connected and running unattended, with the failure paths handled rather than discovered.

WATCH

The AI is supervised, automatically

A second AI grades every answer the coach gives. If a change makes answers worse, it rolls back on its own. And the coach can't touch a rider's plan without a tap to approve the exact change first.

Oversight isn't a slide in my pitch deck. It runs in production, unattended, right now.

TEACH

Plain language, everywhere

Tap any number in the app and the coach explains what it means and why it matters for you. Training jargon is translated, not assumed.

Systems people actually understand get used. Ones they don't get abandoned.

The four tags match the four services on the services page. Same standard, different domain: your version gets built around your operation instead of a bike.

See it live

Alpine Velo has its own site

Alpine Velo is on the App Store now. The product and the full story live at alpinevelo.ai.

Download on the App StoreVisit alpinevelo.ai
Start a project

This is the standard. Want it applied to your operation?

Start a projectSee the approach