langchain-simple-RAG

A simple Retrieval-Augmented Generation (RAG) system built with LangChain, Ollama, and PyPDF. This system allows users to ask questions about PDF documents and receive AI-generated answers based on the document content.

Features

• PDF document loading and processing
• Document chunking for efficient retrieval
• Vector embeddings using Ollama's BGE-M3 model
• Question answering using Llama 3.2 Vision (11B parameters)
• Interactive command-line interface
• Source attribution for answers

Prerequisites

• Python 3.12 or higher
• Ollama installed and running locally
• Required models pulled in Ollama:
  • bge-m3 for embeddings
  • llama3.2-vision:11b for text generation

Installation

1. Clone this repository:

git clone https://github.com/WytheHuang/langchain-simple-RAG.git
cd langchain-simple-RAG

1. Set up a Python virtual and activate it:

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

1. Install the required packages:

uv sync

Usage

1. Create a documents directory in your project root and place your PDF files there:

mkdir documents
cp /path/to/your/pdfs/*.pdf documents/

2. Run the main script:

python main.py

3. Enter your questions when prompted. Type /quit to exit the program.

Example interaction:

Loading documents...
Loaded 25 document chunks

Enter your question (or '/quit' to exit): What is machine learning?

Answer: [AI-generated answer will appear here]

Sources:
Source: documents/example.pdf
Page: 1
Content: [Relevant excerpt from the document]

Project Structure

langchain-simple-RAG/
├── documents/            # Directory for PDF files
├── main.py              # Main application script
├── README.md            # This file
├── pyproject.toml       # Project configuration
└── uv.lock             # Dependencies lock file

How It Works

1. Document Loading: The system loads PDF documents from the documents directory using PyPDFLoader.

2. Document Splitting: Documents are split into smaller chunks using RecursiveCharacterTextSplitter for efficient processing.

3. Embedding: Document chunks are embedded using Ollama's BGE-M3 model and stored in an in-memory vector store.

4. Question Answering: When a user asks a question:

   • The system retrieves relevant document chunks
   • Passes them to the Llama 3.2 model
   • Generates a contextual answer
   • Provides source attribution

Configuration

The system uses the following default settings:

• Chunk size: 1000 characters
• Chunk overlap: 200 characters
• Embedding model: bge-m3
• LLM model: llama3.2-vision:11b

Development

To contribute to this project:

1. Install development dependencies:

uv sync --group lint

2. Run linters:

ruff check .
black .

License

APACHE-2.0 License See the LICENSE file for details.