KP-OCR – Intelligent Document Platform

KP-OCR is a production-ready document intelligence platform designed with cloud constraints, cost efficiency and reliability in mind.

🔗 Live Demo

https://kp-doc-intelligence-gfdabpaqg6hbcybz.canadacentral-01.azurewebsites.net/

Note: Some features require authentication.

🚀 Features

• CNN-based document classification
• OCR-driven text extraction
• Secure PDF invoice generation
• Subscription-based billing
• GST-compliant invoices
• Role-based access control
• Cloud-native deployment (Azure App Service)

🏗️ Architecture

Frontend

• Jinja2 templates
• Bootstrap UI

Backend

• Flask (Python)
• MongoDB (Atlas)
• ONNX Runtime for inference

Document Processing

• CNN-based classifier
• OCR extraction pipeline
• ReportLab-based PDF generation

PDF Generation

• ReportLab (pure Python, no native deps)

Deployment

• Azure App Service (Linux)
• Gunicorn WSGI server

🧠 Key Engineering Decisions

• Replaced TensorFlow with ONNX Runtime in production to significantly reduce memory footprint and cold-start latency.

• Selected ReportLab over HTML-to-PDF tools to avoid native Cairo dependencies that often fail on managed cloud platforms.

• Deferred digital PDF signing (pyHanko) intentionally due to crypto and OpenSSL dependency instability on low-tier instances.

• Designed invoice generation to be fully in-memory, eliminating filesystem dependencies and improving security.

🔐 Security

• All billing routes are authenticated
• Invoice access is user-scoped and validated server-side
• No sensitive keys or credentials are stored in the repository
• PDF generation is stateless and ephemeral

⚖️ Trade-offs & Limitations

• HTML/CSS-based PDF rendering was avoided, which limits pixel-perfect visual styling in favor of reliability.

• Free-tier cloud constraints influenced dependency choices and deferred certain enterprise features.

• The current OCR pipeline prioritizes accuracy and validation over raw throughput.

These trade-offs were made intentionally to ensure stability, cost control and predictable deployments.

📸 Screenshots

The following screenshots demonstrate key user flows from document upload to billing and invoice generation.

Pricing & Plans

User Dashboard

OCR Upload & Processing

OCR Results & Validation

User Profile & Subscription

Billing History

Generated Invoice (PDF)

📌 Status

Actively developed and deployed in a production environment.

Planned enhancements include:

• Digital PDF signatures
• Enterprise compliance workflows
• Advanced analytics and audit logging