Kempt

A document processing app that lets you chat with PDFs, extract text from scanned documents, and generate summaries.

What it does

Chat with documents: Upload a PDF and ask questions about it. The app finds relevant sections and gives you answers based on the actual content.

Fix broken PDFs: Got a scanned PDF where you can't select text? Upload it and get back a proper PDF with searchable text.

Summarize documents: Generate concise summaries of long documents.

Demo

Original Image

Output Image

( This image when received would be in a pdf form with selectable text )

Quick Start

With Docker (easiest)

git clone https://github.com/yourusername/kempt.git
cd kempt/backend

# Add your environment variables to .env file
cp .env.example .env

docker build -t kempt .
docker run -p 8080:8080 --env-file .env kempt

Frontend runs separately:

cd frontend
npm install
npm start

Manual setup

You'll need:

• Node.js 16+
• Python 3.8+
• MongoDB
• Redis
• Ollama with qwen2.5vl model

# Backend
cd backend
npm install
cd python && pip install -r requirements.txt

# Frontend  
cd frontend
npm install

# Start Ollama
ollama serve
ollama pull qwen2.5vl:7b

# Run everything
npm start # in backend
npm start # in frontend

How it works

Document chat: Uses ChromaDB to store document embeddings, then searches for relevant chunks when you ask questions. Google Gemini generates the final answer.

OCR: Converts PDF pages to images, runs them through the Qwen2.5VL vision model to extract text, then creates a new searchable PDF.

Storage: Files go to Google Cloud Storage, user data in MongoDB, sessions in Redis.

Environment setup

Create .env in the backend directory:

# Database
mongoID=mongodb://localhost:27017/kempt
REDIS_HOST=localhost
REDIS_PORT=6379
REDIS_PASSWORD=

# Auth
jwt_secret=your-secret-key
CLIENT_ID=google-oauth-client-id
CLIENT_SECRET=google-oauth-client-secret

# Google Cloud
GCS_BUCKET_NAME=your-bucket
GOOGLE_APPLICATION_CREDENTIALS=path/to/service-account.json
GOOGLE_API_KEY=your-gemini-api-key

# AI
OLLAMA_API=http://localhost:11434/api/generate

PORT=8080

API

# Auth
POST /auth/signup
POST /auth/signin
GET /auth/LoginWithGoogle

# Document processing
POST /user/qa          # Chat with documents
POST /user/fix         # OCR processing  
POST /user/summarize   # Generate summary

Tech stack

• Backend: Node.js, Express
• Frontend: React
• Databases: MongoDB (users), Redis (state storage using oauth), ChromaDB (document vectors)
• AI: Ollama (OCR), Google Gemini (chat), LangChain (document processing)
• Storage: Google Cloud Storage

Limitations

• Free users get 5 requests which when over are renewed every *8hours
• OCR works best with clear, high-quality scans,
• Document chat quality depends on how well the text chunks are created and how clean the question is
• Requires Google Cloud account for file storage (of the OCR file)

Development

# Run tests
npm test

# Format code
npm run format

# Type checking
npm run type-check

Deployment

Works on any platform that supports Docker. Tested on:

• Google Cloud Run (backend + OLLAMA)
• Vercel (frontend)

Contributing

Open an issue first to discuss what you'd like to change. Pull requests welcome.

License

MIT