AiLyze

Inspiration- We've all been there: not being able to understand complex medications, whether we can get a cheaper alternative, forgetting to take our daily vitamins, mixing up complex medication schedules, or worrying if an elderly family member took their evening pills. The consequences of poor medication adherence can be severe. I wanted to build a centralized, smart platform that takes the guesswork out of health management. A tool that not only sends reminders but intelligently parses doctor prescriptions, calculates health scores, and keeps family members connected through an SOS and tracking ecosystem.

How it was built - Frontend: A mobile-first interface built using React Native (Expo). For dynamic and smooth animations, I integrated framer-motion, moti, and gsap. The UI uses lucide-react for beautiful iconography and Tailwind CSS concepts for styling. Backend: A robust REST API powered by FastAPI and SQLAlchemy for database management. AI & OCR Pipeline: When a user uploads a prescription, the app leverages EasyOCR and PyTesseract to extract raw text, which is then semantically parsed using Anthropic's Claude, spaCy, and HuggingFace transformers.

Challenges we ran into -

State Synchronization: One of the biggest hurdles was keeping the frontend Dashboard fully synced with the backend Health Score. Toggling a dose in the DoseTracker required immediate cache-busting and state updates so the user saw their score jump in real-time.
Third-party Integrations: Getting the Twilio SOS OTP feature robust enough to reliably send SMS to emergency contacts involved meticulous error-handling and working around 503 Service Unavailable API errors.
Complex Timezone Logic: Standardizing daily streaks and resets meant shifting from UTC to the user's localized time. Resetting a user's progress at their midnight without contaminating a family member's streak required careful chron/database design.
AI Parsing Accuracy: Raw OCR data from handwritten or blurry prescriptions is incredibly noisy. Orchestrating a pipeline that feeds OCR strings into a deterministic Anthropic LLM prompt to extract structured (Drug Name, Dosage, Frequency) JSON objects took several iterations.

Accomplishments that we're proud of - Successfully building a highly accurate OCR pipeline that integrates PyTesseract and EasyOCR with Anthropic's Claude. It effectively reads and categorizes messy, real-world prescription images into actionable medication regimens without requiring tedious manual entry.

What we learned The main thing we learnt was Machine Learning Integration - Running hybrid lightweight ML models (OpenCV + EasyOCR) alongside API-based LLMs to create a fast, robust data extraction pipeline.

What's next for AiLyze

Voice Input for the Elderly: Adding a voice-to-text feature so elderly users can log doses, check their schedule, or ask questions without needing to navigate complex UI screens.
Real-Life Notification Alert System: Upgrading from basic push reminders to a persistent, real-life notification alert system (like automated phone calls or loud critical alerts) for missed essential medications.
Wearable Device Integration: Syncing health scores and missed dose alerts directly to smartwatches.
Pharmacy Refill Integration: Automatically notifying local pharmacies when a prescription is running low.
Upgrading our database to PostgredSQL for bigger development