// CASE STUDY 03 · SPORTS TECH · AI
Keel
Mental skills coaching for athletes, between sessions with their coach. A 30-second daily check-in, guided routines and an AI weekly summary that helps the coach see patterns, without ever diagnosing.

- 30 sdaily check-in
- 2views: athlete and coach
- 0diagnoses: the AI only drafts, fixed rules check it
01THE CHALLENGE
A mental performance coach sees an athlete for an hour, then loses sight of them until the next session. What happened in training, how they slept, how tense they were before a competition: most of it is forgotten by then.
Health data is sensitive, and an app like this can easily drift into therapy. The tool had to stay inside honest skills coaching and send people to a professional when it is a mental health issue.
02MY ROLE
Product, design and code. I am also studying to become a mental performance coach, so the app follows the scope and referral rules from that training.
- Next.js
- TypeScript
- Supabase
- Vercel AI SDK
- Gemini
- Tailwind
- Vitest
03KEY DECISIONS
- 01
Coaching, not therapy
Athletes read in plain words what the app does and does not do before they start. A "Need help now?" button with 9-8-8 and 911 is on every screen, and set answers lead to a "talk to a professional" screen.
- 02
The AI drafts, rules decide
Patterns are calculated by fixed rules. The AI only writes the summary, and it never sees names, dates or free text. Every sentence must cite the pattern behind it; clinical words or numbers that are not in the data reject the summary.
- 03
Private by default
Athletes join by invitation only and decide what their coach sees. By default the coach sees numbers, not written reflections. Database rules make sure each person reads only what is theirs.
- 04
Skills with a source
Breathing and pre-performance routines cite the research they come from. Anything without a published source is labelled as our own.
- 05
A demo that cannot leak
The public demo uses fictional athletes, never touches the database and does not call the AI live. Anyone can explore both views without an account, with zero risk to real data.
04ON THE PHONE



05RESULTS
A public demo with both views working end to end: check-in, patterns, guided breathing, routines, an example weekly summary and the coach’s timeline.
Automated tests cover the alert rules, the privacy rules and the AI summary checks.
06WHAT’S NEXT
A privacy impact assessment under Quebec’s Law 25, a paid AI plan that does not train on data, and then the first real coach and athletes.
GOT A PROJECT
LIKE THIS?
Tell me what you want to build. I'll reply within 24 hours with straight answers and a fixed quote.
LET'S TALK →