RAG App with Chain of Thought (CoT) and Multi-Query Retrieval For Wordpress website Blog

This project is a Retrieval-Augmented Generation (RAG) application that enhances response accuracy and depth by implementing a Chain of Thought (CoT) approach. By generating multiple alternative queries, the system retrieves a more comprehensive set of documents to answer complex user questions with improved relevance and context.

---

Table of Contents

1. Overview
2. Features
3. Installation
4. Usage
5. How It Works
6. Conclusion
7. Contributing
8. License

---

Overview

This RAG app uses LangChain, a powerful framework for AI applications, to create a system capable of answering user questions with augmented responses. By employing a Chain of Thought (CoT) approach, the app generates multiple interpretations of each question to retrieve related documents from a vector database. The end result is a well-rounded and contextually informed response that mimics human-like reasoning.

Features

• Chain of Thought (CoT) Multi-Query Generation: Generates multiple alternative versions of the user's question to retrieve a broader set of relevant documents.
• Retrieval-Augmented Generation (RAG): Provides accurate answers by retrieving related documents and combining them with the generative model's response.
• Vector Database Integration: Uses Chroma as a vector store to manage document embeddings.
• Web Scraping: Extracts and processes data from web pages for use in the vector database.

Installation

1. Clone the repository:

   git clone /~https://github.com/Mohansharma13/rag-for-wordpress.git

2. Navigate to the project directory:

   cd rag-app-with-cot

3. Install dependencies:

   pip install -r requirements.txt

4. Set up your environment variables:
   • Create a .env file in the root directory.
   • Add your GOOGLE_API_KEY and any other required API keys.

Usage

1. Run the Application: Launch the app using Streamlit:

   streamlit run app.py

2. Interact with the RAG System: Enter a question, and the system will generate multiple interpretations to retrieve the most relevant documents and respond accurately.

How It Works

1. Data Retrieval and Vectorization:

   • Data is retrieved from a specified URL using a custom web scraper.
   • After retrieving the raw content, it's cleaned of HTML tags and irrelevant shortcodes.
   • The cleaned text is converted into document embeddings using sentence-transformers/all-MiniLM-L6-v2, then stored in a Chroma vector database.

2. Chain of Thought (CoT) Multi-Query Generation:

   • For each user query, the app generates multiple alternative queries through a prompt template in LangChain. This multi-query retriever enhances accuracy by capturing various angles of the initial question.

3. Multi-Query Retrieval:

   • The generated queries are used to retrieve documents from the vector database.
   • By combining the results of these queries, the system assembles a richer context for answering the question.

4. Response Generation:

   • The retrieved documents are passed to a generative AI model (e.g., ChatGoogleGenerativeAI) to produce a comprehensive response.
   • If the context is insufficient, the model relies on its training knowledge to generate a complete answer.

Conclusion

The CoT-based multi-query strategy provides a smarter way of handling complex and open-ended questions, allowing the app to deliver responses that are both accurate and contextually enriched. This approach demonstrates the power of retrieval-augmented generation in real-world AI applications.

Contributing

Contributions are welcome! Please fork the repository and submit a pull request with any enhancements or bug fixes.

License

This project is licensed under the MIT License.