MedBridge — Custodian-Mediated Medicine Redistribution

AI-verified, NGO/hospital-mediated matching of unused medicines to patients in need. Built for WeMakeDevs Bharat Builds Tour — Ship It track.

Architecture

Donor uploads photo → browser runs Tesseract.js OCR (client-side, free, no AWS AI service)
                                     ↓
                    Extracted text sent to Lambda (verify_medicine) → pure Python parsing
                                     ↓
                            DynamoDB: MedicineListing
                                     ↓
                        Lambda (match_engine) → DynamoDB: NeedRequest
                                     ↓
                        Lambda (safety_check) → rule-based safety logic
                                     ↓
                            Custodian (NGO/Hospital) confirms handoff

Frontend: single HTML page (with Tesseract.js loaded from CDN) → API Gateway → Lambdas Everything sits behind a Custodian entity (verified NGO/hospital) — no direct donor-to-stranger handoff, mirroring the real, legal SMS Hospital Jaipur Drug Bank model.

Note on AI/ML service choice: this build intentionally avoids Bedrock, Textract, and Rekognition, since new AWS accounts often hit an account-verification hold on these managed AI services that can take hours to clear. Using Tesseract.js (client-side OCR) and rule-based safety logic keeps the whole flow working reliably on any AWS account from the moment it's created — a deliberate, documented engineering trade-off for a time-boxed hackathon, not an oversight.

Prerequisites (Day 1 — do this first)

1. Create an AWS account if you don't have one: https://aws.amazon.com/free
2. Install AWS CLI: pip install awscli --break-system-packages (or use the official installer)
3. Install AWS SAM CLI (makes deploying Lambda+API Gateway+DynamoDB one command): https://docs.aws.amazon.com/serverless-application-model/latest/developerguide/install-sam-cli.html
4. Configure your CLI: aws configure (paste your Access Key + Secret Key + region us-east-1)
5. No AI/ML service setup needed — OCR runs client-side via Tesseract.js, loaded automatically from a CDN in frontend/index.html

What's in this folder

• template.yaml — SAM template: defines your S3 bucket, DynamoDB tables, Lambdas, and API Gateway in one file
• lambda/verify_medicine.py — receives OCR-extracted text from the browser, parses drug name/expiry with pure Python, marks it pending_review if valid
• lambda/custodian.py — lets a custodian (NGO/hospital) list pending listings and approve them, flipping status to available
• lambda/match_engine.py — matches an approved listing to an open need request by drug + location, and confirms handoffs
• lambda/safety_check.py — rule-based logic that flags whether a match may need pharmacist review, based on age/condition
• frontend/index.html — upload + custodian dashboard + matching UI, with Tesseract.js loaded from CDN for client-side OCR

Day 1: Deploy the skeleton

cd medbridge
sam build
sam deploy --guided

Answer the prompts (stack name: medbridge, region: us-east-1, allow SAM to create IAM roles: yes). This creates your DynamoDB tables, S3 bucket, Lambda functions, and API Gateway. Copy the API Gateway URL from the output — you'll paste it into frontend/index.html.

Day 2: Wire and test

1. Open frontend/index.html, replace API_URL_HERE with your API Gateway URL
2. Test uploading a medicine photo — check CloudWatch Logs if it fails (Console → Lambda → your function → Monitor → View logs)
3. Manually insert a test NeedRequest row in DynamoDB console to test matching

Day 3: Deploy frontend + polish

# Amplify Hosting - easiest way to get a public URL for your frontend
npm install -g @aws-amplify/cli
amplify init
amplify add hosting
amplify publish

Day 4: Record demo, submit

Show: upload → auto-verify → match → safety flag → custodian confirm. Mention the SMS Hospital Jaipur precedent explicitly — it shows judges you designed around the real-world constraint instead of ignoring it.

Common beginner issues

• OCR reads garbled text: use a clear, well-lit, front-facing photo — Tesseract.js struggles with angled, blurry, or stylized images
• Lambda timeout: increase Lambda timeout in template.yaml if OCR text parsing takes longer than expected on large inputs
• CORS errors in browser: make sure API Gateway has CORS enabled (already configured in template.yaml)
• "Failed to fetch" errors: run the frontend through a local server (e.g. VS Code's Live Server extension) rather than opening index.html directly as a file:// URL — some browser security restrictions block Tesseract.js's worker scripts under file://