Cost control on AI calls
AI endpoints are rate limited per user (Bucket4j) and repeated answers are cached (Caffeine), so one user can't run up the bill.
Work / AI medicine assistant
MedAssist & PharmaLens
Scan or type a medicine to identify it, check drug interactions, find nearby pharmacies and set intake reminders. AI does real work here, not a chat box on top.
Android appREST API2025
The challenge
People keep strips of medicine without the box and can't tell what a tablet is or whether two medicines are safe together. The app had to recognise a medicine from a photo and keep AI costs and abuse under control.
What was built
- Android app: CameraX scanner, pharmacy map, reminders with intake logging, history
- Spring Boot API with Gemini-powered image recognition and analysis
- Roles for users and verified healthcare providers, email verification, password reset
- Rate limiting and caching on the AI endpoints
Engineering decisions
The parts that make it hold up in real use
Clean Architecture on Android
Data, domain and presentation layers with use cases, Hilt for injection and Room for offline data, so features can change without breaking each other.
Stack
KotlinJetpack ComposeHiltRoomCameraXSpring Boot 3PostgreSQLGemini APIDocker
Need a product built this carefully?
Share a few details and you'll get a reply within 24 hours: honest advice, a rough range, and the next step.