AliFullStack

Here’s a full-stack internal tool built with AliFullStack for the Operations team at GreenEarth Solutions to automate their vendor invoice processing workflow. The tool includes a React-based frontend with a drag-and-drop invoice upload interface, a Node.js backend that processes invoices using OCR and LLM-powered validation, and a PostgreSQL database to store vendor data and invoice statuses. The LLM (using the `mistral-7b-instruct` model) extracts key fields like vendor name, invoice number, and amount, then cross-references them with the company’s vendor database to flag discrepancies.

**Frontend (React):**
- A dashboard displays a real-time table of pending invoices, sorted by urgency. Each row includes a 'Review' button that triggers the LLM validation process.
- A form allows users to manually input invoice details if OCR fails, with dropdowns pre-populated from the vendor database.
- The UI updates dynamically via WebSocket connections to the backend, ensuring the team sees changes in real-time.

**Backend (Node.js):**
- The `/upload` endpoint accepts PDF/JPEG/PNG files, stores them in an S3 bucket, and queues them for processing.
- The `/process` endpoint uses Tesseract.js for OCR and the LLM to extract and validate invoice data against the vendor database. Invalid invoices are flagged for manual review.
- The `/status` endpoint provides a RESTful API for the frontend to fetch invoice statuses and updates.

**Database (PostgreSQL):**
- Schema includes tables for `vendors` (id, name, contact, approved_invoice_threshold), `invoices` (id, vendor_id, file_path, status, extracted_data), and `audit_logs` (id, invoice_id, action, timestamp).

**LLM Integration:**
- The LLM is prompted with: "Extract the vendor name, invoice number, date, and total amount from this invoice. Return the data in JSON format. If the vendor name doesn’t match our database, flag it as invalid."
- The backend parses the LLM’s response and updates the invoice status accordingly.

**Deployment:**
1. Clone the AliFullStack template: `git clone https://github.com/ali-fullstack/ali-fullstack-template.git`
2. Install dependencies: `npm install` (frontend) and `npm install` (backend).
3. Set up the database: `docker-compose up -d` (PostgreSQL + S3 mock).
4. Configure the LLM: Add your API key to `.env` and select the model in `config/llm.js`.
5. Run the app: `npm run dev` (frontend) and `npm start` (backend).
6. Deploy using AliFullStack’s built-in CLI: `ali deploy --env production`.

The tool reduced invoice processing time by 60% in the first week of use, with the LLM handling 85% of validations automatically.