Project Title: Ask Document RAG

Overview

This project provides a sophisticated document retrieval system that leverages the capabilities of the LlamaIndex for efficient query processing. The architecture is designed to operate locally, integrating advanced Retrieval-Augmented Generation (RAG) strategies with powerful language models like Mistral, Gemma, and Meta for natural language processing.

Architecture

The architecture is split into several key components:

1. User Interface (UI)

   • Streamlit: Provides a user-friendly interface for querying documents. Users interact with the system by entering their queries through the Streamlit application.

2. Document Processing and Querying

   • Document Upload: Users can upload documents which are then processed and stored locally.
   • LlamaIndex:
     • Chunking: Documents are split into smaller, manageable chunks.
     • Embedding: Each chunk is embedded into a vector space using LlamaIndex’s vectorization techniques, allowing for efficient searching and retrieval.
     • Advanced RAG Retrieval Strategies: Implements advanced RAG strategies to enhance the retrieval process.
   • Prompt Templates: Custom templates are used to format and present the retrieved documents as context for generating responses.

3. Data Storage

   • PostgreSQL: A PostgreSQL database, running in a Docker container, is used to store the indexed chunks and their embeddings. This enables persistent storage and quick retrieval of document data.

4. Language Model Interaction

   • Colab Machine: A Google Colab instance is used to run the language models. This allows for powerful computation without requiring local resources.
   • Ngrok: Ngrok tunnels are used to securely expose the Colab machine's server to the local environment, facilitating communication between the components.
   • Language Models: Supports multiple models including Mistral AI, Gemma, and Meta, for generating responses based on the queries.

5. Response Generation

   • After retrieving the most relevant chunks from the database, the system formulates a prompt incorporating the retrieved context and sends it to the language model. The model then generates a response, which is relayed back to the user through the Streamlit interface.

Setup and Installation

Prerequisites

• Python 3.x
• Docker
• PostgreSQL
• Streamlit
• Google Colab Account
• Ngrok Account

Installation Steps

1. Clone the Repository

   git clone https://github.com/Elma-dev/ask_document_rag.git
   cd ask-document-rag```

2. Install Dependencies

   pip install -r requirements.txt
   

3. Setup Ngrok for Colab Machine
4. Run the Application

   streamlit run app.py
   

Usage

Upload Document: Use the Streamlit interface to upload documents that you want to query.

Submit Query: Enter your query in the provided input box.

Receive Response: The system will process your query, retrieve the most relevant information from the documents, and generate a response using the language model. The response will be displayed on the Streamlit interface.

Contributing

If you wish to contribute to this project, please fork the repository and submit a pull request with detailed information about the changes you made.