LLM RAG Application

A Retrieval-Augmented Generation (RAG) system that allows you to query PDF documents using Large Language Models. This application uses DeepSeek Chat for natural language processing and OpenRouter API for model access.

Features

• PDF document processing and text extraction
• Efficient document embedding generation
• Vector database for similarity search
• RAG-powered question answering
• Integration with DeepSeek Chat model

Prerequisites

• Bun runtime environment
• OpenRouter API key (or other OpenAI-compatible provider)

Setup

1. Install dependencies:

bun install

2. Set up your OpenAI/OpenRouter API key:
   • Copy the .env.example file to .env:

   cp .env.example .env

   • Edit .env and replace your-api-key-here with your actual OpenAI/OpenRouter API key

Project Structure

• main.ts: Entry point for the application
• pdf.ts: PDF processing and text extraction
• embeddings.ts: Document embedding generation
• vectorDB.ts: Vector database operations
• rag.ts: RAG implementation and query handling

Usage

The application supports two main commands:

1. Ingesting a PDF

To embed a PDF document into the vector database:

bun main.ts ingest <pdf_file>

Example:

bun main.ts ingest AyaMohsenResume.pdf

This will:

• Extract text from the PDF
• Split the text into manageable chunks
• Generate embeddings for each chunk
• Store the chunks and embeddings in the vector database

2. Querying the Embedded Document

To ask questions about the ingested document:

bun main.ts query "<your question>"

Example:

bun main.ts query "Explain who this person is and can they be good for game development?"

The system will:

• Generate an embedding for your question
• Find relevant context from the PDF using similarity search
• Use the DeepSeek Chat model to generate an accurate answer based on the retrieved context

Dependencies

• @xenova/transformers: For generating embeddings
• openai: For interacting with the OpenAI/OpenRouter API
• pdf-parse: For extracting text from PDF documents