Prescription-Decoder---The-Handwriting-MCP-Agent

Inspiration

Illegible doctor handwriting causes thousands of medication errors daily. Pharmacists guess, patients suffer, and lives are at risk. We built a solution to make every prescription crystal clear.

What it does

Transforms messy prescription photos into structured digital data—checks allergies/drug interactions, creates FHIR standards, and texts patients instructions in their language. One photo → safe medication.

How we built it

Python MCP server with GPT-4 Vision for handwriting → FHIR conversion → interaction checking → Twilio SMS. Published as A2A agent on Prompt Opinion Marketplace with Streamlit UI for testing.

Challenges we ran into

Handwriting variability (20+ doctor styles), mapping Indian meds to RxNorm codes, maintaining SHARP context across tool chain, and getting vision model to read cursive consistently.

Accomplishments we're proud of

95% accuracy on test prescriptions, real-time allergy alerts, Hindi SMS working end-to-end, and fully integrated with Prompt Opinion platform in under 48 hours.

What we learned

Vision models + healthcare standards can solve real problems. MCP/A2A makes complex workflows simple. Indian healthcare needs more AI solutions built locally.

What's next

10+ regional languages, pharmacy inventory integration, learning from pharmacist corrections, and pilot with actual hospitals.

Built With

• langchain
• llm