RxScan AI

The Inspiration

Prescription errors are a silent crisis. Even in 2026, many patients struggle to decipher handwritten scripts, leading to missed doses or dangerous drug interactions. I wanted to build a bridge between complex clinical data and accessible, patient-safe information.

How I Built It

RxScan AI is a Multi-Agent RAG (Retrieval-Augmented Generation) pipeline.

OCR Processing: The system takes a prescription image and extracts text using vision-language models.
Data Orchestration: I used LangChain to orchestrate the LLM (Groq) to map extracted text against the NIH RxNav API.
The Hallucination Barrier: To ensure clinical accuracy, the system doesn't just trust the LLM. It performs a secondary "validation layer" where it verifies drug names against official clinical databases. If the confidence score is low, it flags the medication for pharmacist review.

Challenges I Faced

Data Mapping: Mapping unstructured OCR text to structured medical databases like RxNav was non-trivial. I overcame this by implementing a fuzzy-matching logic layer in my API routes.
Latency: Running RAG on medical data can be slow. By using Groq’s LPU inference engine, I reduced response times from seconds to milliseconds, creating a snappy experience even on mobile.

What I Learned

I learned that building AI for high-stakes fields like healthcare is 20% model tuning and 80% data validation. Reliable RAG is all about designing robust guardrails for when the AI is "unsure."