MedTrack Health OS — Personal AI Health Analytics
Role
Full cycle: from idea and architecture to development and deployment.
Tech
iOS: SwiftUI, RevenueCat, HealthKit, OneSignal. Backend: Fastify + TypeScript, PostgreSQL, Docker. AI: OpenAI GPT-4o-mini — document parsing, score generation, RAG chat with health context. Architecture: Server-driven UI, A/B experiments with weighted randomization, 23 push scenarios with timezone/cooldown, CI/CD deployment on push to main.
Key features
- 01Paywall layout
- 02Paywall type
- 03Paywall placement
- 04Onboarding variants
- 05Purchase journey
Problem
People keep health diaries, collect lab results, visit doctors — but the data is scattered across paper notes, PDFs, and chats. Nobody sees the connections between sleep and blood pressure, between medication and wellbeing, between test results and symptoms. A doctor sees a patient for 15 minutes twice a year. The rest of the time, people are on their own — without the tools to understand what's affecting them.
For the business, there's a second problem: health apps lose 80% of users in the first week. Without systematic work on retention and conversion, even a great product doesn't survive.
Audience
- People with chronic conditions — see what triggers flare-ups, arrive at appointments with ready analytics rather than "I think it got worse."
- Health-conscious users — optimize sleep, stress, and activity through data, not intuition.
- Patients after medical tests — store results and reports in one place, track biomarker trends on charts.
What's different
| Manual / Competitors | MedTrack |
|---|---|
| Data spread across notes, Excel, different apps | Everything in one place with AI document parsing |
| "I think it got worse" | Objective trends and correlations from your data |
| Generic advice from the internet | AI assistant that knows your health history |
| Static charts | Dynamic dashboard with scores by organ system |
| Requires discipline and time | 30-second check-in, auto-sync with Apple Health |
| One paywall, hope for the best | A/B tests across layout, type, placement with cohort analytics |
| Push: "Don't forget to open the app" | 23 contextual scenarios with escalation and cooldown |
| Metrics — MAU and DAU | Purchase journey, cohort analysis, feature-to-conversion correlation |
Screenshots
Admin Panel
Customization
- 01Tracking modules — toggled from the backend, new ones can be added without an app release.
- 02Diary fields — server-driven schema: type, validation, order — all managed through the admin panel.
- 03AI model — switchable via settings (gpt-4o-mini → gpt-4o → any other).
- 04Paywall — A/B across 3 axes: layout (3 variants) × type (3 variants) × placement (2 variants) = 18 combinations without a release.
- 05Push scenarios — 23 triggers with configurable cooldown, time windows, and conditions.
- 06Cohort analytics — segmentation across 6 axes: demographics, behavior, time, source, features, subscription.
- 07Units — metric/imperial, switchable at the user level.
Need something like this?
Tell me about the task — I'll suggest an approach, pick a stack and scope it out.