Smart Invoice Processing System — InvoiceVision

A college project (Summer Internship / Computer Engineering, GTU) demonstrating an AI-assisted invoice management SaaS: upload an invoice, have its fields extracted automatically, review AI-generated insights, and move it through an approval workflow — with a dashboard and analytics on top.

This build follows the features and pages described in the project report and presentation (abstract, objectives, database design, UI pages, workflow), but keeps the implementation simple and free to run:

• No paid cloud services required. The report's target architecture uses Azure AI Document Intelligence, Azure OpenAI, Azure SQL Database, Azure Blob Storage, and Azure AD B2C. For this demo, those are replaced with equivalent client-side logic — real text extraction from digitally-generated PDFs (pdfjs-dist) plus real in-browser OCR (tesseract.js) for scanned pages and photos, a template-based "AI insights" generator, and localStorage in place of a database — so the whole app runs entirely in the browser and deploys for free on Vercel. Only a genuinely unreadable file (very low- quality scan, corrupted PDF) falls back to simulated placeholder data, clearly flagged as such in the UI.
• Everything else matches the report: the Draft → Pending → Approved/Rejected → Paid workflow, the dashboard/invoices/upload/analytics/ notifications/profile/settings pages, search & filters, CSV export, and the Azure Portal-inspired UI.

Tech stack

• Next.js 15 (App Router) + TypeScript
• Tailwind CSS
• Recharts (dashboard & analytics charts)
• lucide-react (icons)
• pdfjs-dist (embedded PDF text extraction) + tesseract.js (in-browser OCR)
• Browser localStorage as the data layer (users, invoices, notifications)

Features

• Sign up / sign in (demo auth, stored in localStorage)
• Dashboard with KPI cards, monthly spend trend, and status breakdown
• Drag-and-drop invoice upload (PDF/PNG/JPG) with real field extraction — embedded PDF text when available, OCR (scanned pages/photos) otherwise
• Editable invoice detail page with line items and AI insights (summary, payment recommendation, risk flag, duplicate/anomaly detection)
• Approval workflow: Draft → Pending → Approved/Rejected → Paid
• Invoice list with search, status filter, and CSV export
• Analytics: monthly trend, category & vendor breakdown, paid vs. pending, tax summary
• Notifications feed
• Profile page and a Settings page with a working light/dark mode toggle

Running locally

npm install
npm run dev

Open http://localhost:3000. Sign up with any name/email/ password — a few sample invoices are seeded automatically so the dashboard and analytics aren't empty on first login.

Deployment

Deployed on Vercel — push to main and Vercel builds automatically. No environment variables are required.

Project background

Based on the "Smart Invoice Processing System" project report and summer internship presentation (GTU / SAL College of Engineering, Computer Engineering). See the original report for the full target architecture, database schema, and objectives.