RAG Assited Abstract Question Answering

This repository contains a program built with Pinecone vector storage, Groq, and Streamlit. The program is designed to parse documents, split them into chunks, convert these chunks into vectors, store them in a Pinecone vector database, and allow for efficient querying and retrieval of similar matches and their corresponding metadata. Additionally, a Streamlit frontend has been created to facilitate user interaction with the database.

Showcase

 

Features

• Document Parsing and Chunking: A Jupyter Notebook is used to parse documents and split them into manageable chunks.
• Vector Conversion and Storage: These chunks are then converted into vectors using the model all-mpnet-base-v2 and stored in a Pinecone vector database.
• Efficient Querying: The database can be queried using a search query to retrieve similar matches along with their corresponding metadata.
• Streamlit Frontend: A user-friendly frontend built with Streamlit allows users to query the database and view the sources of the retrieved information.
• LLM Integration: The context from the retrieved data can be passed to an LLM (LLaMA 3-8b-8192) to provide an answer to the user's question using the available context from the vector storage.
• Source Filtering: Optionally, the information to be retrieved can be filtered by its source (e.g., book or article title).

How It Works

1. Document Parsing:

   • Documents are parsed and split into smaller chunks using a Jupyter Notebook.

2. Vector Conversion:

   • The chunks are converted into vectors using the all-mpnet-base-v2 model.

3. Vector Storage:

   • The vectors are stored in a Pinecone vector database.

4. Querying the Database:

   • Users can query the database to retrieve similar matches and their metadata.

5. Streamlit Frontend:

   • The Streamlit frontend allows users to interact with the database and view the results.

6. LLM Integration:

   • The context from the retrieved data is passed to an LLM (LLaMA 3-8b-8192) to provide answers based on the available context.

7. Source Filtering:

   • Users can filter the information to be retrieved by its source (e.g., book or article title).

Installation

To get started, clone this repository and install the necessary dependencies:

git clone https://github.com/yourusername/document-vector-search.git
cd document-vector-search
pip install -r requirements.txt

Usage

1. Update API Keys:

   • Copy .env.example to .env and update your API keys

2. Run the Jupyter Notebook:

   • Place your documents in the articles folder
   • Parse and split your documents into chunks.
   • Convert the chunks into vectors and store them in the Pinecone database.

3. Run the Streamlit Frontend:

   • Start the Streamlit application to query the database and view the results.

streamlit run app.py

Contributing

Contributions are welcome! Please open an issue or submit a pull request for any enhancements or bug fixes.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgements

• Pinecone for the vector database.
• Hugging Face for the all-mpnet-base-v2 model.
• Streamlit for the frontend framework.
• Groq for the computation acceleration.

---

Happy querying!