An AI health platform taken from empty repo to the App Store.
AI health assistant · Web, iOS, Android
The Challenge
Consumer health AI has a credibility problem. Most of it either answers from general knowledge with no view of the individual, or it edges into clinical advice it has no business giving. Valdura needed an assistant grounded in a member's actual bloodwork that could hold context across months of conversation, and it needed to know exactly where its own authority ended.
Our Approach
We treated the safety architecture as the product, not as a layer added at the end. Every AI response runs through automated evaluations before a member sees it, and anything that crosses into clinical judgment routes to that member's own clinician rather than getting answered. The web platform and both native apps came out of one shared codebase, so a change to the safety rails ships everywhere at once instead of three times. An over-the-air update pipeline means most fixes reach members' phones without waiting on a store review.
What We Built
- Next.js web platform and Expo iOS/Android apps from one shared codebase
- Safety rails, clinical-review routing, and automated evals on every AI response
- Approved on the App Store and Google Play, with multiple releases and over-the-air updates shipped since launch
On Mobile
The native app on iOS and Android. Biomarker triage, a body map, sleep and recovery, and the assistant one tap away. Shown with demo data.
Built With
- Next.js
- React Native / Expo
- Neon Postgres
- Drizzle ORM
- Anthropic Claude
- Stripe
- Vercel