Retrieval Augmented Generation (RAG) Chatbot with Streamlit, LangChain, and Pinecone

This project implements a Retrieval Augmented Generation (RAG) chatbot using Streamlit for the UI, LangChain for orchestration, Pinecone for vector storage, HuggingFace embeddings, and a local Large Language Model (LLM) via llama.cpp.

Features

• Document Ingestion: Load and process PDF documents from the documents/ folder
• Vector Storage: Store document embeddings in Pinecone vector database
• Retrieval: Perform similarity search to retrieve relevant context
• Local LLM: Use Mistral 7B Instruct model locally for generating responses
• Chat Interface: Interactive Streamlit app for chatting with the RAG system

Prerequisites

• Python 3.11+
• Pinecone account (free tier available)
• Mistral 7B Instruct GGUF model file

Installation

1. Clone the repository:

   git clone https://github.com/trisita-26/LangChain-Pinecone-RAG.git
   cd LangChain-Pinecone-RAG
   

2. Create a virtual environment:

   python -m venv venv
   

3. Activate the virtual environment:

   • Windows: venv\Scripts\activate
   • macOS/Linux: source venv/bin/activate

4. Install dependencies:

   pip install -r requirements.txt
   

5. Set up accounts and models:

   • Create a free account on Pinecone
   • Download the Mistral 7B Instruct GGUF model from Hugging Face (e.g., mistral-7b-instruct-v0.1.Q4_K_M.gguf)
   • Place the model file in a models/ directory

6. Create a .env file in the root directory with your API keys:

   PINECONE_API_KEY=your_pinecone_api_key_here
   PINECONE_INDEX_NAME=sample-index
   

Usage

1. Ingest Documents

Run the ingestion script to process and store your documents:

python ingestion.py

This will:

• Create a Pinecone index (if it doesn't exist)
• Load PDF files from the documents/ folder
• Split documents into chunks
• Generate embeddings and upload to Pinecone

2. Test Retrieval (Optional)

Test the retrieval system:

python retrieval.py

3. Run the Chatbot

Start the Streamlit app:

streamlit run app.py

Open your browser to the provided URL and start chatting!

Project Structure

• chatbot_rag.py: Main Streamlit app for the chatbot interface
• ingestion.py: Script to ingest documents into Pinecone
• retrieval.py: Script to test document retrieval
• requirements.txt: Python dependencies
• documents/: Folder for PDF documents to ingest
• models/: Folder for local LLM model files (create this directory)

Configuration

• Pinecone Index: Configured for 384 dimensions (matching HuggingFace all-MiniLM-L6-v2 embeddings)
• Chunk Size: 500 characters with 50 character overlap
• Retrieval: Top 3 similar chunks with score threshold of 0.5
• LLM: Mistral 7B Instruct with temperature 0.7, max 512 tokens

Customization

• Change the LLM model path in chatbot_rag.py
• Adjust retrieval parameters in the retriever setup
• Modify the prompt template for different response styles
• Add more document types by updating the loader in ingestion.py

Troubleshooting

• Ensure your Pinecone API key is correct and has sufficient credits
• Check that the model file path matches the one in app.py
• Verify that documents are in PDF format in the documents/ folder
• Make sure the virtual environment is activated before running scripts

License

This project is for educational purposes. Please check the licenses of individual components (LangChain, Pinecone, etc.) for commercial use.

• Rename .env.example to .env
• Add the API keys for Pinecone and OpenAI to the .env file

Executing the scripts

1. Open a terminal in VS Code

2. Execute the following command:

python sample_ingestion.py
python sample_retrieval.py
python ingestion.py
python retrieval.py
streamlit run chatbot_rag.py

LangChain-Chatbot