DocFlow

DocFlow is a full-stack document intelligence platform for uploading files, extracting text, running OCR, chatting with documents, generating summaries, and performing PDF operations from one workspace.

Stack

• Frontend: React, React Router, Zustand, Axios, Tailwind CSS
• Backend: Node.js, Express, Mongoose, Multer
• Database: MongoDB
• Storage: Supabase Storage
• AI/OCR: local AI service, Hugging Face fallback, Tesseract OCR, PDF/DOCX/XLSX/PPTX parsing

Features

• User authentication with JWT
• Upload support for PDF, DOCX, TXT, PNG, JPG, XLSX, and PPTX
• Text extraction and OCR for scanned/image-based documents
• RAG-style document chat
• Key-information document summarization
• PDF compress, merge, and split tools
• Search inside processed document text
• User history and admin dashboard

Project Structure

docfloe/
├── backend/
│   ├── controllers/
│   ├── middleware/
│   ├── models/
│   ├── routes/
│   ├── services/
│   └── server.js
└── frontend/
    ├── public/
    └── src/

How It Works

1. A user signs up or logs in.
2. The frontend stores the JWT and calls protected backend routes.
3. A document is uploaded through the backend using Multer.
4. The backend detects file type and extracts readable text.
5. If needed, OCR is used for scanned PDFs or images.
6. Extracted text is chunked and stored with document metadata.
7. The original file is uploaded to Supabase Storage.
8. AI features use the processed chunks for chat and summary generation.

Backend Setup

From backend/:

npm install
node server.js

Required environment variables in backend/.env:

PORT=5000
MONGODB_URI=your_mongodb_connection_string
JWT_SECRET=your_jwt_secret
SUPABASE_URL=your_supabase_url
SUPABASE_ANON_KEY=your_supabase_anon_key
LOCAL_AI_SERVICE_URL=http://localhost:5001
AI_SERVICE_TOKEN=your_internal_ai_service_token
HF_API_KEY=your_huggingface_api_key
HF_FALLBACK_MODEL=your_huggingface_model
FRONTEND_URL=http://localhost:3000

Optional local AI service:

python ai_service.py

Set the same AI_SERVICE_TOKEN for the Node backend and the local Python service. The backend first tries the local AI service and then falls back to Hugging Face if needed.

Frontend Setup

From frontend/:

npm install
npm start

If needed, define:

REACT_APP_API_URL=http://localhost:5000/api

Main API Areas

• /api/auth for register, login, and current user
• /api/documents for upload, list, search, and delete
• /api/ai for summarize, chat, logs, and summary download
• /api/pdf for compress, merge, split, and document fetch
• /api/admin for admin stats, user management, and document management

Current AI Flow

• Upload pipeline extracts and normalizes document text
• Text is split into chunks
• Chat retrieves relevant chunks and answers from that context
• Summarization retrieves key chunks and generates a concise structured summary
• Fallback logic is used only when AI output is weak or unavailable

Notes

• Uploaded files are temporarily stored in backend/uploads/ before processing
• Supabase stores the original or generated file
• MongoDB stores document metadata, extracted text, chunk data, chat history, and summaries

Scripts

Backend:

npm run dev
npm run start
npm run ai-service

Frontend:

npm start
npm run build