Eczcalibur

I have eczema. The hardest part isn't the itch — it's that by the time you get to the dermatologist, you can't remember what triggered the last three flares. For children, it's worse: clinical action plans are written for adults, kids don't follow them, and parents fly blind between appointments.

Eczcalibur is what I wish existed when I was a kid.

What it does

Eczcalibur bridges the gap between dermatologist appointments for children aged 7–14 and their parents.

The child uses a Zelda-style interface to log flare-ups — picking their mood, marking affected body areas, and optionally uploading a photo. Every log earns points toward a prize store their parent manages. Compliance feels like a quest, not a chore.

The parent gets:

• A real-time dashboard showing zone history (Green / Yellow / Red) and flare trends
• A Claude-generated Written Action Plan built on the NEA WAP framework — personalized to the child's specific triggers, affected areas, and medications
• An AI chat that answers questions like "what's been triggering flares this month?" using the child's actual log history

How we built it

• Expo React Native — single codebase across iOS and Android
• Supabase — PostgreSQL with row-level security (scoped per Clerk user), real-time sync across devices
• FastAPI on Fly.io — backend handling all Claude API calls with Clerk JWT auth
• Claude Opus — plan generation, parent chat, and pre-appointment clinical summaries
• Claude Code — used throughout as an AI pair programmer; all architectural decisions reviewed and owned by us

Challenges

Getting a child to voluntarily log health data is a UX problem before it's a technical one. The gamification layer had to feel genuinely rewarding, not like a form dressed up with coins. Balancing that with a clinical-grade parent interface — same data, completely different presentation — was the core design tension of the whole build.

On the technical side: real-time two-device sync required careful token sequencing (Clerk JWT had to be injected before Supabase hydration, not after), and Claude's action plan needed guardrails to stay grounded in the NEA framework rather than generating free-form medical advice.

What we learned

AI in pediatric health is only ethical if the child has agency. The gamification isn't decoration — it's the consent mechanism. The child logs because they want to, not because they're being monitored.

Built With

• api
• asyncstorage
• claude
• clerk
• expo.io
• fastapi
• fly.io
• native
• open-meteo
• postgresql
• python
• react
• router
• supabase
• typescript
• zustand