Enhanced Retrieval-Augmented Generation (RAG) System

Advanced document Q&A with semantic search, reranking, and hybrid retrieval.

Features

• Hybrid dense/sparse retrieval
• Semantic chunking
• Query expansion
• Reranking
• Document compression
• GPU/CPU support

Quick Start

1. Clone the repository

   git clone https://github.com/feyzollahi/SimpleRAG.git
   cd SimpleRAG

2. Install dependencies

   poetry install

3. Configure environment variables

   Create a .env file in the project root (see below for example).

   HF_TOKEN=your_huggingface_token
   MODEL_NAME=google/gemma-2b-it
   EMBEDDING_MODEL_NAME=BAAI/bge-small-en-v1.5
   RERANKER_MODEL_NAME=BAAI/bge-reranker-base
   DOC_DIR=docs
   CACHE_DIR=.cache
   

   • HF_TOKEN is required for some Hugging Face models. Get it from Hugging Face.
   • Adjust other variables as needed.

4. Add your documents

   Place .txt, .md, or .pdf files in the docs directory.

5. Run the app

   streamlit run app.py

.env Example

HF_TOKEN=your_huggingface_token
MODEL_NAME=google/gemma-2b-it
EMBEDDING_MODEL_NAME=BAAI/bge-small-en-v1.5
RERANKER_MODEL_NAME=BAAI/bge-reranker-base
DOC_DIR=docs
CACHE_DIR=.cache

Usage

• Enter your question in the app UI.
• Adjust retrieval and generation settings in the sidebar.
• View sources and context for each answer.

Notes

• For best results, use high-quality documents.
• GPU is recommended for faster inference.
• All configuration is managed via .env.

License

MIT

Acknowledgements

• LangChain
• Hugging Face
• Streamlit