Dynamic Quoting System for Complex Products Using Multi-Agent AI Architecture

Overview

This project implements a sophisticated multi-agent system for generating personalized, context-aware quotes for complex products and services. It leverages LangChain/LangGraph to create an intelligent quoting workflow that incorporates:

• Retrieval-Augmented Generation (RAG) for context-sensitive data retrieval
• Multi-Agent Architecture with specialized roles
• Intelligent Classification using sentiment analysis
• Dynamic Workflow Management through a state graph system
• Database Integration for storing category rates

Prerequisites

Required Python Packages

pip install langgraph pip install langchain-core pip install langchain-openai pip install langchain-groq pip install langchain-community pip install python-dotenv pip install pydantic pip install typing-extensions pip install chromadb pip install langchain-text-splitters

Environment Variables

Create a .env file with the following:

OPENAI_API_KEY=your_openai_key
LANGCHAIN_API_KEY=your_langchain_key
GROQ_API_KEY=your_groq_key
LANGCHAIN_TRACING_V2=optional
LANGCHAIN_PROJECT=optional

Project Structure

Core Components

1. Database Setup

   • SQLite database for category rates
   • Predefined categories with associated rates
   • Automated table creation and data population

2. Assistants

   • Main Assistant: Guides users through information gathering
   • Underwriting Assistant: Evaluates risk and validates categories
   • Quote Assistant: Calculates and presents final premiums

3. Agents

   • Retriever Agent: Extracts and summarizes business operations
   • Reasoning Agent: Determines relevant insurance categories
   • Classification Grading Agent: Evaluates category assignments
   • Quote Generation Agent: Calculates final premiums

4. State Management

   • MainState: Core workflow state
   • RAGState: Retrieval state
   • ExtraState: Additional workflow data

5. Routing Nodes

   • route_main_assistant: Manages main assistant flow
   • route_underwriting_assistant: Handles underwriting transitions
   • route_quote_assistant: Controls quote generation flow
   • route_to_workflow: Determines assistant routing
   • Additional utility nodes for state updates and message handling

Usage

Running the System

1. Open the Jupyter notebook (contextual_quoting_agentic_system.ipynb)
2. Run all cells in sequence
3. The system will prompt you for input when ready
4. Type your questions/responses when prompted
5. Type '/exit' to end the session

Key Features

1. Dynamic Category Classification

• Intelligent matching of business descriptions to categories
• Grade-based evaluation of category relevance
• Contextual understanding of business operations

2. Automated Quote Generation

• Revenue-based premium calculation
• Multi-category rate application
• Dynamic rate adjustment based on business complexity

3. Workflow Management

• State-based conversation flow
• Error handling and fallback mechanisms
• Conditional routing between assistants

Contact

Hector -> hector@hetoll.com