Skip to content

Latest commit

 

History

History
123 lines (86 loc) · 7.99 KB

README.md

File metadata and controls

123 lines (86 loc) · 7.99 KB

System Architecture and Design

Overview

This application provides a chatbot interface designed for Admins and Users. Admins configure prompts that guide the chatbot's responses, while Users interact with the chatbot to gain specific insights based on the local knowledge base.

An advanced feature, Retrieval-Augmented Generation (RAG), enables the chatbot to retrieve and generate answers based on user-uploaded documents, providing responses grounded in specific knowledge.

System Roles and Responsibilities

  • Admin: Configures and manages prompts through the admin interface to ensure that the chatbot provides relevant and accurate responses.
  • Users: Queries the chatbot with industry-specific questions, uploads files or images for knowledge enrichment, and receives structured responses.

System Architecture Diagram

Components and Interactions

  1. Admin Interface:

    • Admins configure prompts through the admin terminal, which are stored in a dedicated SQLite database (Prompt Library).

  2. Prompt Library:

    • A SQLite database holding all configured prompts, accessible only to the backend server for efficient prompt retrieval and updates.

  3. Prompt Engine:

    • Retrieves relevant prompts from the Prompt Library based on user queries.
    • Processes user-uploaded files and images, combining this input with stored prompts to create an API request for the chatbot.

  4. Retrieval-Augmented Generation (RAG) Module using LangChain:

    • Implements RAG to leverage user-uploaded documents as contextual knowledge sources.
    • Document Ingestion: Supports various file formats (e.g., PDF, DOCX, TXT) and extracts relevant knowledge from these documents using LangChain’s document loader.
    • Embedding and Retrieval: Uses LangChain to create embeddings of the extracted text and stores them in a FAISS vector database.
    • Disk-Based Persistence: FAISS’s save_local and load_local methods ensure that embeddings are saved across application restarts.
    • Contextual Query Retrieval: When a query is received, LangChain retrieves relevant information from the uploaded documents using similarity search and integrates it into the prompt.
    • This module allows the chatbot to generate answers that are based on specific knowledge provided by the user, improving accuracy and relevance.

  5. LLM (Large Language Model):

    • The Prompt Engine sends API requests to OpenAI's LLM (GPT-4.0), which processes combined prompts and retrieved information from the RAG module.
    • The LLM generates responses using both the pre-configured prompts and user-specific data as context.

  6. Visualization Engine:

    • Formats the response from the LLM to ensure compatibility with the front-end display.

  7. Chatbot Interface:

    • Displays responses to the Users in a conversational format and supports interactions like file uploads and image processing.

Workflow of Capability Components

Screenshot 2024-11-23 at 2 31 30 pm

  1. Prompt Setup by Admin:

    • The Admin configures prompts through the admin interface, and these are stored in the Prompt Library.

  2. User Query with File Upload:

    • Users submit queries and can upload files containing industry-specific information. The Prompt Engine and RAG module process these inputs.

  3. Processing and RAG-Enhanced Response Generation:

    • The Prompt Engine retrieves relevant prompts and combines them with user-uploaded data.
    • The RAG module extracts and retrieves specific content from the uploaded files to enhance the context of the API request.
    • Generated embeddings for document chunks are stored in a FAISS vector store for similarity-based retrieval.
    • The enriched request is sent to the LLM, which generates a response based on both stored prompts and specific user knowledge.

  4. Response Delivery:

    • The chatbot interface displays the response, enriched by the RAG module, providing the user with an answer that reflects the uploaded document's context.

Technology Stack

Screenshot 2024-11-23 at 2 33 53 pm

Front-End

  • React JS: Used for building dynamic user interfaces with a component-based structure for reusability.
  • PrimeReact: A React component library providing pre-built, styled components to streamline UI development.
  • DOMPurify: Used for sanitizing inputs and preventing XSS attacks, ensuring secure rendering of text content.

Back-End

  • Django: Serves as the primary framework, offering ORM, authentication, form processing, and other web application capabilities.
  • Django REST Framework (DRF): Facilitates API development, enabling efficient communication between front-end and back-end.
  • SQLite: A lightweight database for prompt storage during development, with an option to scale to PostgreSQL or MySQL for production.
  • Swagger: Provides interactive API documentation, allowing easy API testing and debugging.
  • LangChain: Integrates with OpenAI for prompt handling and context management, and also supports RAG by enabling document ingestion, embedding, and retrieval.
  • OpenAI (GPT-4): A language model that processes complex queries, generating responses based on configured prompts and user-specific data.

DevOps

  • Docker: Containerizes the application for consistency across different environments, simplifying deployment and scaling.
  • Git: Manages source code version control, supporting team collaboration and maintaining code integrity.

User Interface Design

Chat Page

  • Key Features:

    • Text Interaction: Supports direct text conversation with the chatbot.
    • File Upload: Allows Users to upload files (e.g., PDFs, DOCX, TXT) as knowledge sources for RAG.
    • RAG-Enhanced Responses: Generates answers based on both configured prompts and user-provided documents.

  • Enhanced UX Features:

    • Auto-Scroll: Automatically scrolls to the latest response.
    • Chat History Storage: Retains chat history in local storage for reference.
    • Upload Indicators: Displays a loading indicator during processing and notifies users of unsupported file types.

Admin Page

  • Primary Functions:
    • Prompt Configuration: Admins can view, edit, and set default prompts to guide the chatbot’s behavior.
    • Prompt Management: Allows efficient management of prompt content to improve chatbot responses.
    • Authentication (Future Work): Plans to add identity verification for secure access to the admin interface.

Key Back-End Features

  • Modular Architecture: The backend uses a modular design, enabling independent development and maintenance of functional components.
  • Efficient Data Management: Optimized for handling diverse data types and relationships, with indexing for improved query performance.
  • Stable API Services: DRF enables stable API interfaces, supporting reliable communication between the front-end and back-end.

Chatbot Capabilities

Screenshot 2024-11-23 at 2 33 02 pm
  • Prompt Engine: Manages prompt retrieval, processing, and input-output handling.
  • RAG with LangChain: Integrates LangChain’s document loader, embedding, and retrieval capabilities to enhance responses using user-uploaded documents.
  • LangChain Integration: Handles prompt chaining and context retention, ensuring relevant responses based on past interactions.
  • Multi-Modal Input Processing: Accepts and processes different types of inputs (text, files, images) to deliver tailored responses.
  • Adaptive System Prompting: Default and system prompts define the chatbot’s behavior, guiding its responses based on query type.
  • Contextual Retrieval Setup: Configured similarity search within the FAISS vector store to retrieve relevant content based on user queries.