RAG CLI Tool

A command-line interface tool that implements Retrieval-Augmented Generation (RAG) using Google's Gemini API. This tool allows you to ask questions about the content of text, PDF documents and Python Codebases.

Features

• Process single documents or entire directories
• Supports .txt, .pdf, .py files
• Uses Google's Gemini API for text generation
• Implements RAG pattern using FAISS vector store
• Interactive Q&A interface
• Ability to mention number of relevant chunks to extract depending upon the specifity of answer required (directly proportional)

Prerequisites

• Python 3.8 or higher
• Google API key with access to Gemini API

Installation

1. Clone the repository
2. Create a virtual environment:

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

3. Install dependencies:

   pip install langchain langchain-google-genai python-dotenv faiss-cpu pypdf

4. Create a .env file in the project root:

   GOOGLE_API_KEY=your_api_key_here
   

Usage

Processing a Single File

python rag_cli.py path/to/your/document.txt

Processing a Directory

python rag_cli.py path/to/your/directory

Interactive Mode

After processing the document(s), you can:

• Type your questions about the document content or codebase
• Get AI-generated answers based on the document context
• Type 'quit' or 'exit' to end the session

How It Works

1. Document Processing:

   • Loads documents using LangChain's document loaders
   • Splits documents into manageable chunks

2. Vector Store Creation:

   • Creates embeddings using Google's embedding model
   • Stores embeddings in a FAISS vector store

3. Query Processing:

   • Retrieves relevant document chunks for each query
   • Generates contextual answers using Gemini API

Error Handling

The tool includes comprehensive error handling for:

• Invalid API keys
• Missing files or directories
• Unsupported file types
• Password-protected PDFs
• API rate limits

Limitations

• PDF processing ignores images
• Requires active internet connection
• Subject to Google API rate limits