dosed

Demystifying Prescriptions

Comment 1

Inspiration

Managing medications shouldn't be this hard. People struggle with:

  • Unreadable prescription labels
  • Awkward doctor visits where they forget to ask questions
  • Confusing insurance coverage
  • Drug interactions they don't understand
  • Existing apps that suck

We built DOSED for people who can't afford expensive healthcare or don't have easy access to doctors for quick questions.

What it does

DOSED scans your medication bottles with AI, tracks your prescriptions, checks for drug interactions, and provides an intelligent health assistant.

Key features:

  • AI bottle scanning - just take a picture
  • Prescription tracking - automatically organized
  • Drug interaction checking - stay safe
  • Smart health assistant - answers your questions
  • Doctor finder - locate providers who take your insurance

How we built it

Built as a React Native mobile app with three AI agents:

OCR Agent

Takes picture → OCR.space API → Rule-based extraction → Medication name
  • Uses OCR.space API for text recognition
  • Custom algorithm extracts drug names from prescription labels

Research Agent

Medication query → Vector search → DrugBank API → Interaction results
  • Kaggle datasets + NCBI PubMed for medication database
  • DrugBank API for drug interaction checking
  • Vector similarity search for accurate results

Communication Agent

User question → Gemini API → Contextual response → Discord reminders
  • Gemini 1.5 Flash for intelligent responses
  • Discord integration for medication reminders
  • Provider search using healthcare APIs

Tech Stack:

  • Storage: Supabase PostgreSQL + AsyncStorage
  • AI: Gemini 1.5 Flash + NVIDIA Agent IQ
  • Frontend: React Native with Expo

Challenges we ran into

  • OCR reliability - medication bottles have blurry, rotated text
  • Name extraction - distinguishing drug names from pharmacy labels
  • AI responses - making the chatbot helpful instead of saying "ask your doctor" to everything
  • API integration - managing multiple healthcare APIs with different requirements
  • Data syncing - keeping phone and cloud database in sync

What we're proud of

  • AI that knows your medications and gives personalized advice
  • Reliable scanning even with poor image quality
  • Natural chatbot that feels helpful, not robotic
  • 50,000+ medication database with comprehensive drug info
  • Seamless integration between OCR, AI, and database systems

What we learned

  • AI integration requires extensive fine-tuning
  • You need multiple backup plans when APIs fail
  • Users want AI that actually helps, not constant doctor referrals
  • Sometimes simple algorithms beat complex ML for specific tasks
  • Healthcare APIs have complex restrictions and rate limits

What's next for DOSED

  • Zocdoc Integration - direct appointment booking
  • More Languages - expanding beyond English and Spanish
  • Better AI - smarter medication and health responses
  • Smart Reminders - intelligent medication notifications
  • Insurance Integration - better coverage checking and affordable options

Built With

+ 1 more
Share this project:

Updates