RAG (Retrieval-Augmented Generation) System

A powerful RAG system that allows you to query documents using local or cloud-based language models. This system supports both Ollama (local) and Google AI (cloud) models for generating answers based on your indexed documents.

🌟 Features

Core Functionality

• Document Indexing: Scrape and index web pages with configurable crawl depth
• File Upload: Upload and index PDF, TXT, and Markdown files directly
• Hybrid Search: Combines vector similarity (60%) with keyword matching (40%) for better results
• Conversational Memory: Chat with your documents with follow-up question support
• Multiple Model Support:
  • Local models via Ollama (qwen3:4b, llama3.1:8b, mistral:7b, gemma2:9b, phi3:medium, qwen2.5:7b)
  • Cloud models via Google AI (Gemini 2.5 Flash)

User Interface

• Modern Chat Interface: Ask questions in a natural conversation flow
• Real-time Scraping Progress: Live progress bar, ETA, and logs during web scraping
• Markdown Rendering: Answers and search results display with proper formatting
• Dark Mode: Full dark theme support
• Glassmorphism UI: Modern, premium design with smooth animations
• Toast Notifications: Elegant feedback for all actions

🚀 Quick Start

Prerequisites

• Python 3.8+
• Ollama installed and running (for local models)
• Google AI API key (for Gemini 2.5 Flash)

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/rag-system.git
   cd rag-system

2. Install dependencies:

   pip install -r requirements.txt

3. Start the application:

   python app.py

4. Open your browser and navigate to http://localhost:5000

🛠️ Usage

Adding Sources

Web Scraping:

1. Go to "Add Source" → "Web Scraper" tab
2. Enter a URL, optional name, crawl depth, and max pages
3. Click "Start Scraping" and watch real-time progress
4. Content is automatically chunked and indexed

File Upload:

1. Go to "Add Source" → "File Upload" tab
2. Drag & drop or click to upload PDF, TXT, or Markdown files
3. Files are automatically processed and indexed

Searching & Asking Questions

Semantic Search:

• Go to the "Search" section
• Enter your query and see results ranked by hybrid score (vector + keyword)

Chat with Documents:

• Use the Dashboard chat interface
• Ask questions and get AI-generated answers with source citations
• Ask follow-up questions - the system remembers context
• Click "New Chat" to start a fresh conversation

Settings

• Choose between Ollama (local) or Google AI (cloud)
• For Google AI, enter your API key
• Test the connection before saving

🔧 Configuration

Environment Variables

Create a .env file in the project root:

GOOGLE_API_KEY=your_google_ai_api_key
EMBEDDING_MODEL=google/embeddinggemma-300m
LLM_MODEL=qwen3:4b

📂 Project Structure

rag-system/
├── app.py                # Main Flask application
├── requirements.txt      # Python dependencies
├── static/
│   ├── css/
│   │   └── style.css     # Custom styles
│   └── js/
│       └── main.js       # Frontend logic
├── templates/
│   └── index.html        # Main application UI
└── data/
    └── vectors.db        # SQLite database with vector storage

🤖 API Endpoints

Endpoint	Method	Description
/api/scrape	POST	Scrape URL with SSE progress streaming
/api/upload	POST	Upload and index a file (PDF, TXT, MD)
/api/search	POST	Hybrid search (vector + keyword)
/api/answer	POST	Generate answer with conversational context
/api/stats	GET	Get document statistics
/api/delete-source	POST	Remove a source and its chunks
/api/test-model	POST	Test model connection

📦 Dependencies

• Flask - Web framework
• sentence-transformers - Text embeddings
• sqlite-vec - Vector storage in SQLite
• ollama - Local LLM interface
• google-generativeai - Gemini API
• beautifulsoup4 - Web scraping
• PyMuPDF - PDF text extraction
• marked.js - Markdown rendering (frontend)

🙏 Acknowledgments

• Ollama for providing easy access to local LLMs
• Google AI for the Gemini models
• Sentence Transformers for text embeddings
• SQLite Vector for efficient vector storage