TraceCheck AI

TraceCheck AI is a hackathon prototype for pharma incoming-material verification. It helps receiving or QA teams compare a delivery note, certificate of analysis, and material label before releasing a batch into production.

Repo structure

• frontend/ contains the Vite + React product UI
• backend/ contains the Express API and the Azure Functions app
• shared/ contains the TraceCheck domain types and validation logic used by the frontend and Express API

What is live

• Multi-page React + TypeScript frontend built with Vite
• Shared routed product shell with landing, features, and workspace pages
• Node + TypeScript API layer for secure Azure integration
• API-backed field extraction for uploaded documents
• Editable field-review workflow with OCR baseline restore and approval state
• Cross-document validation and mismatch detection
• Risk scoring and release / manual review / hold recommendation
• Exportable verification report

Azure integration layer

• GET /api/integration/status reports whether Azure Document Intelligence is configured
• POST /api/documents/extract extracts text and fields from uploaded documents
• POST /api/analysis returns the current TraceCheck decision payload
• Binary image and PDF uploads use Azure Document Intelligence when credentials are configured
• Text uploads stay fast by using direct API-side parsing plus TraceCheck field extraction

Azure Functions deployment structure

• A standalone Azure Functions v4 TypeScript app now lives in backend/function-app/
• It mirrors the same backend routes as the local Express API:
  • GET /api/integration/status
  • POST /api/documents/extract
  • POST /api/analysis
• Use this when you want a real Azure deployment target instead of the local development server

What is mocked or fallback-only

• OCR for binary files when Azure credentials are missing
• ERP and supplier master-data integration
• Historical analytics and approval workflow

Getting started

npm install
npm run dev

This starts both:

• the frontend in frontend/
• the Express API in backend/ on port 8787

If you only want the frontend:

npm run dev:frontend

If you only want the Express backend:

npm run dev:backend

If you want the Azure Function App instead:

npm run dev:functions

Build every app:

npm run build

Preview the frontend bundle:

npm run preview

Run the split app with Docker Compose:

npm run docker:up

That stack exposes:

• frontend on http://127.0.0.1:4173
• backend on http://127.0.0.1:8787

Azure setup

Copy backend/.env.example to backend/.env and fill in:

AZURE_DOCUMENT_INTELLIGENCE_ENDPOINT=
AZURE_DOCUMENT_INTELLIGENCE_KEY=
AZURE_DOCUMENT_INTELLIGENCE_MODEL_ID=prebuilt-read
PORT=8787

Copy frontend/.env.example to frontend/.env when you want to override the default API wiring:

TRACECHECK_DEV_API_TARGET=http://127.0.0.1:8787
VITE_TRACECHECK_API_BASE_URL=

Without these values, the app still works in fallback mode and the API status endpoint will report that Azure is not configured.

Frontend wiring options:

• Default local Express API:
  • Keep TRACECHECK_DEV_API_TARGET=http://127.0.0.1:8787
• Docker Compose preview proxy:
  • Set TRACECHECK_PREVIEW_API_TARGET=http://backend:8787 inside the frontend container
• Local Azure Functions host:
  • Set TRACECHECK_DEV_API_TARGET=http://127.0.0.1:7071
• Deployed Azure Function App:
  • Set VITE_TRACECHECK_API_BASE_URL=https://<your-function-app>.azurewebsites.net
  • The frontend will call the same /api/... routes against that base URL

If the frontend is served from a different domain than the deployed Function App, configure Azure CORS on the Function App to allow the frontend origin.

Docker compose

The root docker-compose.yml is set up for the split repo structure:

• frontend builds the Vite app, serves it with vite preview, and proxies /api to the backend container
• backend runs the Express API directly from the split backend/ workspace
• Azure credentials remain optional, so the compose stack still boots in fallback mode without extra secrets

Optional Azure values can be injected into Docker Compose from your shell or a root .env file:

AZURE_DOCUMENT_INTELLIGENCE_ENDPOINT=
AZURE_DOCUMENT_INTELLIGENCE_KEY=
AZURE_DOCUMENT_INTELLIGENCE_MODEL_ID=prebuilt-read

Useful commands:

• npm run docker:build
• npm run docker:up
• npm run docker:up:detached
• npm run docker:down
• npm run docker:logs

Operational flow

1. Open the landing page and navigate to the workspace.
2. Upload the three incoming material documents.
3. Review the extracted fields and correct OCR misses if needed.
4. Confirm the validation matrix and flagged discrepancies.
5. Approve reviewed documents and export the verification report.
6. Use the Azure status card to explain the deployment path.

Suggested Azure architecture

• Azure Static Web Apps for hosting
• Azure Functions or a serverless API layer for validation orchestration
• Azure AI Document Intelligence for structured document extraction
• Azure Vision for OCR on material-label images
• Azure Language for plain-language explanation or summarization

Notes

• The current app is intentionally honest about what is live and what remains roadmap work.
• The Azure integration layer is safe by default: when credentials are missing or a request fails, TraceCheck falls back instead of breaking the workflow.