50% of runners get injured each year. Most are training alone off a generic PDF, with no one to interpret or personalize it for them. At the same time, they have more wearable data than ever — pace, heart rate, sleep — but little of it gets translated into how to actually improve performance and recovery.
Solution
An AI coach that builds personalized programming, interprets your data — including wearable metrics — and knows when to speak up, when to ask how you're feeling, and when to adjust your plan accordingly.
Key Touchpoints
Onboarding
Most training apps ask a runner to answer a dozen questions before they see any value. Achilles gets them from account creation to their first real interaction fast, saving deeper personalization for once they're already in motion.
Training Hub: From a Generic PDF to a Real Plan
Most runners already have a plan — a coach's PDF, a template — they just can't act on it day to day. The Training Hub parses what they upload into a structured week, with today's workout front and center.
Daily Insights: Feedback That Actually Goes Somewhere
Rating a workout shouldn't be a one-way action. Every rating feeds directly into the next insight or adjustment, so a runner can see their feedback actually shape their plan, not disappear into a log no one reads.
Coach: A Place to Ask, Not Just Be Told
Insights and plans only go so far without room for a runner to ask why, or push back. Coach is where a runner can question a recommendation, report how they're feeling, or dig into a data point directly — the same way they would with a real coach.
Stats: Progress You Can Actually See
Personal bests, training volume, and countdown-to-race context, all in one view — so improvement isn't just implied by the plan working, it's visible.
Impact
50
runners onboarded across NYC run club beta
30%
of beta users returned weekly for insights and recommendations
How I Got Here
Initial brainstorming
Where We Started: Voice AI Coaching
We initially considered voice AI coaching, thinking it could create a more interactive experience closer to an in-person coach — and Apple announcing Workout Buddy (WWDC 2025) partway through our build seemed to back that up, though even Apple framed it as motivation, not real coaching.
Animation Explorations
But our initial survey discovered: runners didn't want an AI talking to them mid-run, disruptive regardless of framing. What they actually wanted was insight around how to improve performance/recovery after the workout.
Considering the Anatomy of an Insight
Every insight follows the same structure: a signal in the data, domain knowledge connecting it to a cause, and a recommendation tied to both. Slower pace on a hot day isn't a fitness drop, it's a heat effect. A heart rate spike after three hard days isn't random, it's fatigue — but only when there's enough signal to say so with confidence. When there isn't, Achilles doesn't guess. It asks.
Finding the Right Testers
We went to NYC run clubs directly and recruited runners fitting the exact problem: training for a race, on a generic plan, no coach. That specificity mattered — testing with runners who already had a coach wouldn't have told us whether the coaching held up. 50 signed up for early access and ran a TestFlight beta with real training data — real paces, weather, fatigue, not simulated scenarios.
Edge Case: Designing for Runners Without Wearables
Not every runner trains with a smartwatch — for many first-time racers, a phone is the only device they have. That meant a chunk of Achilles's richest signal (heart rate, HRV, sleep) didn't exist for them. Rather than guess at fatigue from pace alone, we leaned on what was real: a quick RPE rating and a "how did that feel" after each run. Not as rich as sensor data, but enough for some insights — while staying honest about what wasn't available without a wearable.
Reflections
Reflections
Data is only useful when it's interpreted.
The design work wasn't adding more numbers — it was figuring out which signal mattered right now, for this person, and saying it in a way that felt like a coach, not a report.
The system needs to know its limits.
An AI that guesses is worse than one that asks. Every time Achilles didn't have enough context, we designed it to open a conversation instead of fill in the blank. That restraint is what makes the difference between a tool runners trust and one they ignore.
Where I'd take it next.
Some injuries — the ones runners downplay — need a human in the loop. The next step is connecting runners with real coaches when the system flags something serious enough to warrant it. Not replacing the AI layer, but knowing when to escalate beyond it.
This is just a snapshot of the entire design process.
Reach out to mitaelan07@gmail.com for the full story!
Ideate-Atlas2026 – Present
Founding Designer
Ideate-Sage2025 – Present
Founding Designer
Achilles2025
Founding Designer
Ampcontrol2024
Product Design
Apple2021
Under NDA
PTPAL2024
Product Design
.png?v=20260909b)
