Pill Identification and Drug Safety Chatbot

system

## Inspiration
Every year, 7,000+ Americans die from medication errors, with look-alike/sound-alike (LASA) drugs causing 1 in 3 mistakes. As a team with family members impacted by prescription errors, we built Pillmate to transform smartphone cameras into life-saving tools, combining FDA expertise with cutting-edge AI to prevent tragedies before they happen.

## What it does
Pillmate lets users snap photos of unidentified pills to:

Instantly Identify Medications via AWS-powered imprint analysis
Flag Dangerous Look-Alikes using real-time FDA data cross-checks
Explain Drug Safety through an AI chatbot that answers dosage/interaction questions

## How we built it

AI Vision Core: AWS Rekognition processes pill images → Python/Flask extracts text
Drug Intelligence: Redis-cached FDA API data + BeautifulSoup web scraping
Safety Layer: OpenAI analyzes LASA risks in detected medications
Frontend: React web app with image upload and chat interface

## Challenges we ran into

Imprint Accuracy: Struggled with low-light/angled pill photos
API Limits: FDA's rate limits forced creative Redis caching strategies
Chat Hallucinations: Implemented strict prompt engineering to keep AI responses medically conservative
Limited data related to LASA: Absence of curated LASA dataset with risk levels

## What we learned