RAG (Retrieval-Augmented Generation) Project

This project demonstrates a simple Retrieval-Augmented Generation (RAG) pipeline using Node.js, LangChain, OpenAI, and Qdrant. The goal is to index PDF documents, store their embeddings in a vector database (Qdrant), and perform semantic search and question answering over the indexed content.

---

Table of Contents

• Project Structure
• Setup Instructions
• Environment Variables
• Code Explanation
  • indexing.js
  • query.js
• Docker Compose
• How It Works
• Troubleshooting
• License

---

Project Structure

rag/
├── .env
├── .gitignore
├── docker-compose.yml
├── indexing.js
├── query.js
├── package.json
├── package-lock.json
└── (PDF files to index)

---

Setup Instructions

1. Clone the repository:

   git clone https://github.com/Samrat880/RAG-Assignment.git
   cd rag

2. Install dependencies:

   npm install

   Installation

Install all required packages by running:

npm install @langchain/community @langchain/core @langchain/openai pdf-parse qdrant-client dotenv

Packages used:

• @langchain/community – LangChain community integrations
• @langchain/core – LangChain core utilities
• @langchain/openai – OpenAI integration for LangChain
• pdf-parse – PDF text extraction
• qdrant-client – Qdrant vector database client
• dotenv – Loads environment variables from .env file

3. Set up environment variables:

   • Create a .env file in the rag directory.

4. Start Qdrant (Vector Database) using Docker:

   docker-compose up -d

5. Add your PDF files to the rag directory or specify their path in indexing.js.

6. Index your documents:

   node indexing.js

7. Query your documents:

   node query.js

---

Environment Variables

Create a .env file in the rag directory with the following content:

OPENAI_API_KEY=your_openai_api_key_here
QDRANT_URL=http://localhost:6333

• OPENAI_API_KEY: Your OpenAI API key for embedding and language model calls.
• QDRANT_URL: URL for your local Qdrant instance (default: http://localhost:6333).

---

Code Explanation

indexing.js

This script reads PDF files, splits them into chunks, generates embeddings using OpenAI, and stores them in Qdrant.

Key Steps:

1. Load Environment Variables:
   Uses dotenv to load API keys and configuration.

2. Read PDF Files:
   Uses pdf-parse to extract text from PDF files.

3. Chunk Text:
   Splits the extracted text into manageable chunks for embedding.

4. Generate Embeddings:
   Uses OpenAI's embedding API (via LangChain) to convert text chunks into vector representations.

5. Store in Qdrant:
   Connects to Qdrant and stores each chunk's embedding with metadata (e.g., file name, chunk index).

Example Code Snippet:

// Load environment variables
require('dotenv').config();

// Read and parse PDF
const pdf = require('pdf-parse');
const fs = require('fs');
const dataBuffer = fs.readFileSync('yourfile.pdf');
const data = await pdf(dataBuffer);

// Chunk text and generate embeddings
// ... (chunking logic)
// ... (embedding logic)

// Store in Qdrant
// ... (Qdrant client logic)

---

query.js

This script allows you to query the indexed documents using natural language.

Key Steps:

1. Load Environment Variables:
   Loads API keys and configuration.

2. Accept User Query:
   Takes a question from the user (via command line or hardcoded).

3. Generate Query Embedding:
   Converts the user query into an embedding using OpenAI.

4. Search Qdrant:
   Finds the most similar document chunks in Qdrant.

5. Generate Answer:
   Uses OpenAI's language model to generate an answer based on the retrieved chunks.

Example Code Snippet:

// Load environment variables
require('dotenv').config();

// Accept user query
const query = "What is the main topic of the document?";

// Generate embedding and search Qdrant
// ... (embedding and search logic)

// Generate answer using OpenAI
// ... (completion logic)

---

Docker Compose

The docker-compose.yml file is used to run Qdrant locally:

version: "3.8"
services:
  qdrant:
    image: qdrant/qdrant
    ports:
      - "6333:6333"

Start Qdrant with:

docker-compose up -d

---

How It Works

1. Indexing:

   • PDF files are read and split into chunks.
   • Each chunk is embedded using OpenAI and stored in Qdrant with metadata.

2. Querying:

   • User submits a question.
   • The question is embedded and used to search for similar chunks in Qdrant.
   • The most relevant chunks are passed to OpenAI's language model to generate a final answer.

---

Troubleshooting

• Missing API Key:
  Ensure your .env file contains a valid OPENAI_API_KEY.

• Qdrant Connection Issues:
  Make sure Qdrant is running (docker-compose up -d) and QDRANT_URL is correct.

• Dependency Conflicts:
  If you see npm errors, try installing with --legacy-peer-deps:

  npm install --legacy-peer-deps

---

License

This project is for educational purposes.