RAGStackDemo

Streamlit App URL: https://ragstackdemo.streamlit.app/

Philosopher Quote Retriever with Langchain

This repository contains code for a retrieval system that utilizes Langchain to retrieve quotes from a philosopher based on user queries. The system leverages the following components:

• Langchain: A framework for building conversational AI applications.
• Google Generative AI: A large language model for text generation and retrieval.
• Astra DB Vector Store: A database for storing and searching vector representations of text.

Architecture Overview

Project Structure:

• src/: Contains the core Python code for the chatbot functionality.
  • rag.py: Defines the RAG logic for retrieving quotes based on user queries.
  • chat.py: Implements the chat interaction logic using Streamlit components.
  • app.py: The main Streamlit application file that ties everything together.

Code:

rag.py:

This file handles the core retrieval functionality using Langchain:

• It defines two retrieval chains:

  • rag_chain_with_source_multi: Uses a multi-query retriever for retrieving quotes, leveraging paraphrasing to improve retrieval effectiveness.
  • rag_chain_with_source_basic: Falls back to a vector score retriever if the multi-query retriever fails.

• It utilizes an LLM (large language model) to generate variations of the user's question for better retrieval.

• Environment Variables:

  • GOOGLE_API_KEY: API key for accessing Google Generative AI.
  • ASTRA_DB_API_ENDPOINT: Endpoint for Astra DB.
  • ASTRA_DB_APPLICATION_TOKEN: Authentication token for Astra DB.
  • ASTRA_DB_KEYSPACE_Philo: Keyspace name in Astra DB where quotes are stored.

• LLM (Large Language Model): An instance of GoogleGenerativeAI is created for text generation and retrieval tasks.

• Vector Embeddings: A GoogleGenerativeAIEmbeddings instance is used to generate vector representations of text for efficient retrieval.

• Astra DB Vector Store: A connection to the Astra DB vector store is established using AstraDBVectorStore.

• Retrievers:

  • retriever_vector_score: Retrieves quotes based on vector similarity using the Astra DB vector store.
  • retriever_self_query: Uses the LLM to generate multiple query variations based on the user's question to improve retrieval effectiveness.

• Output Parser: LineListOutputParser parses the LLM response and extracts a list of generated queries.

• Prompts:

  • QUERY_PROMPT: LLM prompt template for generating multiple query variations.

• LLM Chain: Combines the LLM with the QUERY_PROMPT and LineListOutputParser for query variation generation.

• Multi-Query Retriever: retriever_multi_query combines the vector score retriever and the LLM chain for retrieving quotes using both similarity search and paraphrased queries.

• RAG Chain: Defines the core conversation flow:

  • Retrieves context (quotes) from previous interactions.
  • Processes the user's question along with the retrieved context.
  • Uses the LLM to generate a response based on the context and question.
  • Formats the response for the user.

• Runnable Chains:

  • rag_chain_with_source_multi: Uses the multi-query retriever for retrieving quotes.
  • rag_chain_with_source_basic: Uses the vector score retriever for retrieving quotes (fallback if multi-query retriever fails).

chat.py:

This file implements the interactive chat interface using Streamlit:

• It defines the chatHere function to handle user input and generate assistant responses.
• It displays chat history, accepts user input, and displays the assistant's response.
• It utilizes the rag_chain_with_source_multi and rag_chain_with_source_basic functions from rag.py for retrieving responses.
• It stores conversation history in the Streamlit session state.

app.py:

This file is the main entry point for the Streamlit application:

• It sets the app title and user information section.
• It calls the chatHere function from chat.py to handle the chat interaction.

Running the Project:

1. Install dependencies: pip install -r requirements.txt
2. Set the required environment variables (refer to the code for details).
3. Run the script: streamlit run src/app.py
4. The script will open streamlit app and asksto prompt you for a question (e.g., "give a quote by plato").
5. It will then retrieve relevant quotes and provide a response based on the retrieved context.

Note: This is a basic example and can be extended to include additional features such as:

• Support for retrieving quotes by multiple philosophers.
• Filtering quotes based on tags or categories.
• Summarization of retrieved quotes.
• Integration with Evaluation API for feedback and improvement.
• Integration with much better retreival Stratergies like Colbert, and RAPTOR.

Screenshots:

https://ragstackdemo.streamlit.app/