FinRAG

Financial RAG System with Pinecone and Gemma

Overview

This project implements a Retrieval-Augmented Generation (RAG) system for financial documents, combining vector search with a large language model to provide accurate answers to financial queries.

Technologies & Tools

Category	Technologies
Language
Vector DB
NLP
ML
LLM
Framework
Data

Features

• Document processing and chunking
• Vector embedding generation.
• Efficient vector storage/retrieval
• Context-aware question answering
• Automatic model selection based on GPUs

Setup Instructions

• Prerequisites
• Python 3.9+
• GPU with sufficient VRAM (minimum 5GB recommended)
• Pinecone API key
• Hugging Face account (for Gemma model access)

Setup

git clone https://github.com/Shegun93/FinRAG.git
cd FinRAG

Installing Dependencies

pip install -r requirements.txt

Configuration

Create a .env configuration files

PINECONE_API_KEY = "your-api-key"
PINECONE_ENVIRONMENT="region"

Authenticate Hugging Face:

huggingface-cli login

Usage

# Example query
query = "What was the operating profit increase from 2011-2012?"
answer = ask(query)
print(answer)

The system will:

• Retrieve the most relevant document chunks
• Generate accurate answers using the Gemma LLM
• Display both the answer and the context used

Customization

• Chunk size: Adjust the chunk_size parameter in the split_into_chunks function
• Model selection: The system automatically selects the appropriate Gemma model based on available GPU memory
• Temperature: Control answer creativity via the temperature parameter in the ask function

Troubleshooting

• GPU Memory Errors: If you encounter memory issues, try:
• Using the 2B model instead of 7B
• Enabling
• Reducing the max_new_tokens parameter

Pinecone Issues: Ensure your:

• API key is correct
• Index name is unique
• Region matches your Pinecone configuration

License

This project is licensed under the MIT License

Acknowledgments

• Pinecone
• Huggingface
• Google for the Gemma models