SimpleChatWithPDF

A Retrieval-Augmented Generation (RAG) application that enables conversational Q&A with PDF documents using LangChain and OpenAI.

Features

• PDF Document Loading: Load and process multiple PDF files
• Vector Database: Store document embeddings in FAISS for efficient similarity search
• Conversational Memory: Maintains chat history for context-aware responses
• RAG Pipeline: Combines document retrieval with LLM-powered answers

Prerequisites

• Python 3.8+
• OpenAI API key

Installation

1. Clone the repository

   cd SimpleChatWithPDF

2. Install dependencies

   pip install langchain langchain-core langchain-community langchain-openai langchain-text-splitters faiss-cpu python-dotenv pypdf

3. Set up environment variables

   Create a .env file in the project root:

   OPENAI_API_KEY=your_openai_api_key_here

Usage

Step 1: Prepare Your PDFs

Place your PDF files in the ./data/ folder. By default, the project includes:

• azure-cloud-cheatsheet.pdf
• azure-security-best-practices-cheatsheet.pdf

To use your own PDFs, edit saveLoadedPDFToVectorDB.py and update the loaders list:

loaders = [
    PyPDFLoader("./data/your-document.pdf"),
    PyPDFLoader("./data/another-document.pdf")
]

Step 2: Build the Vector Database

Run the script to process PDFs and create the FAISS vector database:

python saveLoadedPDFToVectorDB.py

This will:

• Load all specified PDF files
• Split documents into chunks (1000 chars with 150 overlap)
• Generate embeddings using OpenAI's text-embedding-3-small model
• Save the vector database to ./data/chat_database/

Step 3: Chat with Your Documents

Run the chat script to ask questions:

python chat.py

To ask different questions, edit chat.py and modify:

question = "Your question here"
second_question = "Your follow-up question"

Project Structure

SimpleChatWithPDF/
├── chat.py                      # Main chat application with RAG
├── saveLoadedPDFToVectorDB.py   # Script to build vector database
├── data/
│   ├── azure-cloud-cheatsheet.pdf
│   ├── azure-security-best-practices-cheatsheet.pdf
│   └── chat_database/           # FAISS vector store (generated)
│       ├── index.faiss
│       └── index.pkl
├── .env                         # Environment variables (create this)
└── README.md

How It Works

1. Document Ingestion (saveLoadedPDFToVectorDB.py)

   • PDFs are loaded using PyPDFLoader
   • Text is split into chunks using CharacterTextSplitter
   • Chunks are embedded using OpenAI embeddings
   • Embeddings are stored in a FAISS vector database

2. Conversational Retrieval (chat.py)

   • User questions are reformulated using chat history for context
   • Relevant document chunks are retrieved via similarity search
   • Retrieved context + question are sent to GPT-3.5-turbo
   • Response is generated and chat history is updated

Configuration

Parameter	Location	Default	Description
chunk_size	saveLoadedPDFToVectorDB.py	1000	Size of text chunks
chunk_overlap	saveLoadedPDFToVectorDB.py	150	Overlap between chunks
k	chat.py	6	Number of documents to retrieve
model	chat.py	gpt-3.5-turbo-0125	LLM model for responses

Troubleshooting

Windows: ModuleNotFoundError: No module named 'pwd'

Use direct imports to bypass the problematic package-level import:

from langchain_community.document_loaders.pdf import PyPDFLoader

KeyError: '__fields_set__' when loading database

Delete and regenerate the vector database:

Remove-Item -Recurse -Force .\data\chat_database
python .\saveLoadedPDFToVectorDB.py

Version conflicts

Ensure all LangChain packages are compatible:

pip install -U langchain langchain-core langchain-community langchain-openai langchain-text-splitters

License

MIT