ChatwithURL

AI-Powered Documentation Retrieval Assistant

An intelligent documentation retrieval system that uses LLMs (Large Language Models) and vector search to provide accurate answers to questions about any web content or documentation.

Features

• URL Content Processing: Extract and process content from any web URL
• Intelligent Text Chunking: Splits content into meaningful chunks while preserving context
• Vector Search: Uses Chroma DB for efficient similarity search
• LLM Integration: Powered by Ollama for natural language understanding
• Web Interface: Simple and intuitive UI for asking questions
• Session Management: Maintains separate contexts for different users

Prerequisites

• Python 3.10 or higher
• Ollama server running locally or remotely
• Required Python packages (see Installation)

Installation

1. Clone the repository:

cd ChatwithURL

2. Create and activate a virtual environment:

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

3. Install the required packages:

pip install -r requirements.txt

Project Structure

.
├── main.py              # FastAPI application entry point
├── utils.py          # Core processing functions
├── static/
│   └── index.html      # Web interface
├── requirements.txt    # Project dependencies
└── README.md

Configuration

1. Set up your Ollama server URL or add you LLM in in utils.py:

base_url = 'https://your-ollama-server.com'  

2. Configure the vector store directory in utils.py:

VECTORSTORE_DIR = "path/to/vectorstore"  

Usage

1. Start the FastAPI server:

python main.py

2. Open your web browser and navigate to:

http://localhost:8000

3. In the web interface:
   • Enter a URL to process
   • Wait for content processing to complete
   • Ask questions about the content
   • Receive AI-powered answers based on the content

API Endpoints

• POST /set_url/: Process a new URL

  {
    "url": "https://example.com",
    "session_id": "unique-session-id"
  }

• POST /ask_question/: Ask a question about the processed content

  {
    "question": "What is this article about?",
    "session_id": "unique-session-id"
  }

Web Interface

The web interface provides:

• URL input field
• Question input field
• Response display area
• Session management
• Error handling and status messages

Error Handling

The system handles various error cases:

• Invalid URLs
• Unreachable content
• Processing failures
• LLM server issues
• Session management errors

Technical Details

Content Processing

• Fetches URL and extract the content
• Split the text using RecursiveCharacterTextSplitter
• Preserves document structure

Vector Search

• Uses Chroma DB for vector storage
• HuggingFace embeddings (all-MiniLM-L6-v2)
• Efficient similarity search
• Persistent storage

LLM Integration

• Uses Chatbedrock for natural language processing
• Context-aware responses
• Source attribution
• Confidence scoring

Acknowledgments

• Anthropic model for LLM capabilities
• LangChain for the chain infrastructure
• ChromaDB for vector storage
• HuggingFace for embeddings