🧠 From Clueless to Clarity: Building a Document Intelligence System (and Growing Through It)

“You don’t need to know everything to start. You need to start to know everything.” – that’s what this project taught me.

Over the past few days, I built a full-stack Document Categorization System using Django REST Framework and React — a project that pushed me, challenged me, and ultimately helped me grow in ways no tutorial ever could.

🔍 What the System Does

✅ Secure Uploads – Users can upload documents (PDFs or images).

✅ Text Extraction – Using OCR (via pytesseract) and PDF parsers, the system extracts raw text.

✅ Auto Categorization – A rule-based keyword engine analyzes content and classifies documents as:

• Tax
• Identity
• Medical
• Real Estate
• Other

✅ Confidence Score – Each classification comes with a score to indicate reliability.

✅ JWT Authentication – Built secure access using JWT via SimpleJWT.

✅ Live Filtering + Search – React frontend allows users to search and filter documents dynamically.

✅ Manual Category Override – Users can correct the category if needed.

🔧 Tech Stack

• Backend: Django, DRF, PostgreSQL, JWT Auth, OCR (Tesseract), PDFMiner
• Frontend: React, Axios, Tailwind
• Hosting (local/dev): Postman, VSCode, Railway for DB

🌱 What I Learned (The Real Win)

🧠 Real-world debugging – Nothing teaches better than a 404(server not found )or 500 (Internal server error ) at 2AM.

🧠 React hooks & state management – From total confusion to clarity.

🧠 API design – Including custom endpoints like /update_category/ and understanding router behavior.

🧠 Django internals – Signals, model methods, serializers, request/response cycles.

🧠 Confidence – More than the score I attached to documents, I built some for myself.

💬 A Word to Anyone Starting Out

I didn’t feel “ready” when I started. I felt like an impostor midway. I finished it anyway — because growth doesn’t come from knowing, it comes from doing.

🧩 Core Logic – Behind the Scenes

At the heart of the project lies a simple but effective rule-based classification engine:

1. Text Extraction - PDFs are parsed using pdfminer . Images (JPEG, PNG) are converted to text via OCR using pytesseract.
2. Keyword-Based Categorization Each document is scanned for predefined keywords. For example: "income tax", "gst" → Tax "aadhar", "passport" → Identity "diagnosis", "prescription" → Medical
3. Model Save Logic On saving a document, the backend: 1. Extracts the text , 2. Predicts the category , 3. Stores it with metadata like upload time, user, and confidence.
4. Frontend Integration React handles: 1. Upload , 2. Real-time feedback (category & confidence) , 3. Filtering documents by category, name, or upload time

This simple rule-engine can later evolve into a full-fledged ML-based classifier, but it does the job today — fast, clean, and explainable.

🙌 Open to Feedback, Opportunities & Collaboration

If you're working in AI, OCR, or backend-heavy systems — I’d love to connect, contribute, and grow. And if you’re hiring for backend/full-stack roles — I’m actively looking.