DocuGeniusRAG

Welcome to DocuGeniusRAG, an innovative platform designed to revolutionize how we interact with documents through AI-powered Q&A capabilities. This project leverages the latest advancements in Retrieval-Augmented Generation (RAG) and Natural Language Processing (NLP) to provide users with precise answers to document-related questions, enhancing understanding and accessibility.

Features

• Interactive Q&A with Documents: Upload your PDF documents and ask any questions directly.
• AI-Powered Insights:Get accurate answers powered by state-of-the-art NLP technologies.
• Reference Cross-Checking: Every answer includes reference to the text snippets used, promoting transparency and trust in the AI's responses.

Getting Started

To get started, clone this repository and follow the setup instructions provided in the documentation:

git clone https://github.com/yourrepository/DocuGeniusRAG.git
cd DocuGeniusRAG
pip install -r requirements.txt

Note: Copy config.example.yaml to config.yaml and modify it with your local settings:

Environment Setup

After installing the necessary packages, you need to set up your environment variables to ensure the application can communicate securely with the Block Entropy API.

Configure API Access

1. Obtain and API Key: You must have an API key to interact with the Block Entropy API. If you do not have one, you can request it by signing up at: Block Entropy AI.
2. Setup Your Environment Variables:
   • Create a .env file in the root directory of the project.

   touch .env
   

   • Open the .env file and add your Block Entropy API key as follows:

   BE_API_KEY=your_block_entropy_api_key_here
   

   • Save this file. This file should not be committed to your version control system. Ensure .env is listed in your .gitignore file to prevent it from being uploaded.

Go into web-ui directory and run the app.py script to launch the webserver.

cd DocuGeniusRAG/web-ui
python app.py

After running the server, you can use the web app in your localhost. The home page looks like this:

Usage

After setting up, you can start interacting with your documents:

1. Upload Your Documents: Currently supports PDF format.
2. Ask Questions: Enter your question in the provided interface.
3. Receive Answers: Get answers along with references to the document sections that provided the information.

The links direct you to the page and highlights the text chunks used to generate the particular answer.

Data Flow Diagram

The Data Flow Diagram (DFD) provides a visual representation of the data processing within our Retrieval-Augmented Generation (RAG) application. This diagram helps to illustrate the flow of data between the application's components and how these components interact to process user queries, retrieve relevant information, and generate responses. Understanding the data flow is crucial for both using the application effectively and contributing to its development.

Detailed Walkthrough

1. User Interaction:

   • Users interact with the application through a web-based interface where they can submit the document and interact in a chat space with the LLM.
   • Document are processed as follows:
     • Chunk up the text: The text within these documents is divided into manageable pieces or "chunks". This process is essential for handling large texts and improving the efficiency of the search and retrieval operations.
     • Embeddings: Each text chunk is converted into a vector representation using embedding techniques. These embeddings capture the semantic meaning of the text and are used for matching queries with document content.
     • ChromaDB: The embeddings are stored in ChromaDB, a database optimized for handling vector data, which facilitates quick retrieval based on vector similarity.

2. Retrieve and Rank:

   • In this stage, the system processes queries by finding the most relevant document chunks.
   • Process:
     • Query: This is the user’s input or question that the system needs to respond to.
     • Get k nearest embeddings: The system retrieves the 'k' nearest embeddings from ChromaDB, which correspond to the chunks most relevant to the query. The default settings (k=30 and distance=cosine) indicate that it retrieves 30 closest matches based on cosine similarity.
     • bge-reranker: This component reranks the retrieved embeddings based on BGE-RERANKER model to ensure the best matches are chosen for response generation. Top M=10 chunks are selected for prompt creation.

3. Response Generation:

   • This is the final stage where the system uses the retrieved information to generate a coherent response.
   • Process:
     • Construct prompt with contextual information and query: The system constructs a prompt that includes both the user's query and contextual information derived from the matched document chunks. This step is crucial for ensuring that the generated response is contextually relevant.
     • Block Entropy AI: This is a component handling LLM tasks that takes the constructed prompt and generates a textual response.
     • Generated output along with reference chunks: The final output includes not only the answer to the user's query but also references to the chunks of text from which the information was derived. This adds a layer of transparency and allows users to see the source of the information in the response.

4. Delivering the Response:

   • The generated response is sent back to the user interface, where it is displayed to the user as system chat and the references are provided as link which when clicked opens the page of file with highlighted chunks that had been used.

This flow ensures that each user query is handled efficiently and accurately, providing a seamless experience from query to response.

Contribution

Contributions are welcome! If you have ideas or find bugs, please open an issue or submit a pull request.

License

DocuGeniusRAG is made available under the Apache License 2.0.