RX-Check

AI-powered medication safety and clarity — for every patient, every prescription.

Comment

🧠 Inspiration

As patients or caregivers, it’s easy to get overwhelmed by prescription information: what’s active, what’s expired, what not to mix, and when to refill. This complexity grows even more for older adults or families managing multiple medications across multiple people. We wanted to build a simple, AI-powered tool to make post-prescription care safer, clearer, and more manageable — especially for patients leaving the hospital or managing chronic conditions at home.

💊 What it does

RX-Check is a smart medication management assistant that helps patients and their families:

  • Upload and track active prescriptions
  • Automatically generate safe medication summaries with potential conflicts using Google Gemini
  • Search for drugs semantically using Pinecone embeddings
  • Maintain profiles for multiple family members
  • View interactive medication cards with hover summaries and refill info

All of this happens in a clean, modern UI — no medical degree required.

⚙️ How we built it

  • Frontend: React + Material UI, with Axios for HTTP requests
  • Backend: FastAPI, handling file uploads, profile storage, and API logic
  • AI: Google Gemini (gemini-1.5-pro) used to generate human-friendly medication summaries and conflict warnings
  • Search: Pinecone + Sentence Transformers (all-MiniLM-L6-v2) to enable fuzzy or semantic drug search
  • Database: MongoDB (via Motor) to store user profiles, prescription documents, and relationships
  • File handling: Prescription PDFs uploaded via FastAPI and simulated parsing into prescription entries

We securely call Gemini from the backend, format prompts using drug metadata, and return rich HTML responses to the frontend.

🧱 Challenges we ran into

  • Pinecone's vector schema required careful formatting of metadata to support both exact and semantic drug queries.
  • Gemini outputs needed to be carefully constrained with a structured prompt to avoid hallucinations or verbose responses.
  • Managing active vs. inactive prescriptions dynamically and associating them with user-uploaded files and Mongo documents required schema rethinking mid-hackathon.
  • Avoiding security risks (like exposing the Gemini API key) while still generating responses quickly from the frontend.

🏆 Accomplishments we're proud of

  • Seamless integration of Google Gemini to generate clinical-grade medication summaries.
  • Real-time drug interaction insights embedded directly in prescription cards.
  • Built a working semantic drug search engine in a healthcare context — with fuzzy matching and fallback to LLM summaries.
  • Created a usable multi-user system with support for family members and refill tracking.

📚 What we learned

  • How to fine-tune prompts to control the tone, format, and accuracy of LLM output for medical use cases.
  • Using Pinecone effectively to combine exact-match queries with semantic fallback.
  • Building secure AI-powered apps where sensitive logic stays on the backend.
  • How to unify multiple components (vector search, PDF upload, AI summarization, UI state) into a cohesive, intuitive user experience.

🚀 What’s next for RX-Check

  • Integrate OCR to extract medication names from real prescription labels or pill bottles (image → text → summary).
  • Expand Gemini summaries with real-time FDA interaction data or patient-specific contraindications.
  • Add SMS/email refill reminders and daily schedule notifications.
  • Deploy with patient authentication and connect to real EHR systems using FHIR.

Built With

Share this project:

Updates