PDF Question-Answering System with Vector Search

A production-ready system that enables intelligent question-answering capabilities for PDF documents using OpenAI and vector search technology. Built with security and multi-tenancy in mind.

📖 Read the detailed guide on Medium

Features

• 🚀 Intelligent PDF document processing and analysis
• 🔍 Semantic search using vector embeddings
• 💬 AI-powered question answering using OpenAI
• 🔒 Multi-tenant architecture with team isolation
• 🎯 Production-ready with Docker support

Quick Start

Prerequisites

• Python 3.9+
• Docker and Docker Compose
• OpenAI API access (Azure or regular)
• Qdrant vector database

Installation

1. Clone the repository:

git clone https://github.com/doganarif/pdf-gpt-vectordb-qa.git
cd pdf-gpt-vectordb-qa

2. Create and configure your environment file:

cp .env.example .env
# Edit .env with your credentials

3. Start the services:

docker compose up --build

Usage

1. Upload a PDF document:

curl -X POST http://localhost:8000/upload \
  -F "file=@/path/to/document.pdf" \
  -F "team_id=your_team_id" \
  -F "document_id=doc123"

2. Ask questions about the document:

curl -X POST http://localhost:8000/answer \
  -H "Content-Type: application/json" \
  -d '{
    "team_id": "your_team_id",
    "question": "What is the main topic of the document?"
  }'

Architecture

The system consists of several key components:

• PDF Processing: Extracts and chunks text from PDFs
• Vector Search: Enables semantic document search using Qdrant
• Answer Generation: Utilizes OpenAI for intelligent responses
• API Layer: Provides RESTful endpoints with security measures

Configuration

Key environment variables:

# OpenAI/Azure Configuration
AZURE_OPENAI_ENDPOINT=your_endpoint
AZURE_OPENAI_API_KEY=your_key
AZURE_DEPLOYMENT_NAME=your_deployment

# Qdrant Configuration
QDRANT_HOST=qdrant
QDRANT_PORT=6333

# Application Configuration
DEBUG=false
ENABLE_HTTPS=true
MAX_UPLOAD_SIZE_MB=16

API Endpoints

Endpoint	Method	Description
/upload	POST	Upload and process a PDF document
/answer	POST	Get answers to questions about documents
/documents	GET	List available documents for a team
/health	GET	Check system health status

Security Features

• Team-based isolation for multi-tenant setups
• Rate limiting per team
• Secure file handling
• Authorization middleware
• Input validation

Development

1. Create a virtual environment:

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

2. Install dependencies:

pip install -r requirements.txt

Deployment

The system is containerized and can be deployed using Docker Compose:

# Production deployment
docker compose -f docker-compose.prod.yml up -d

For production deployments, consider:

• Configuring proper authentication
• Implementing backup strategies
• Monitoring and logging

Acknowledgments

• OpenAI for their powerful language models
• Qdrant team for the vector database
• All contributors and supporters

---

Built with ❤️ by doganarif