Automated Order Management System

A LangGraph-based order management system that can handle order placement and cancellation using AI agents. This system demonstrates how to build complex, multi-step workflows with Large Language Models (LLMs) using LangGraph.

Based on Kshitij Kutumbe's article LangGraph AI agents : Building a Dynamic Order Management System : A Step-by-Step Tutorial.

Use Cases

• Intelligent query categorization (Place Order vs Cancel Order)
• Inventory availability checking
• Dynamic shipping cost calculation based on location and weight
• Payment processing simulation
• Order cancellation handling
• State management across the workflow
• Conditional branching based on user intent

Prerequisites

• Python 3.8+
• OpenAI API key (GPT-4 Turbo access required)

Project Structure

ai-driven-order-management/
├── data/
│   ├── inventory.csv      # Sample inventory data
│   └── customers.csv      # Sample customer data
├── src/
│   ├── __init__.py
│   ├── config.py         # Shared configuration and LLM setup
│   ├── main.py           # Entry point
│   ├── tools.py          # LangChain tools
│   ├── nodes.py          # Workflow nodes
│   ├── state.py          # State definitions
│   └── workflow.py       # Workflow graph definition
├── requirements.txt
├── setup.py
└── README.md

Setup

1. Clone the repository:

git clone https://github.com/schmitech/ai-driven-order-management.git
cd ai-driven-order-management

2. Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Create a .env file in the root directory and add your OpenAI API key (copy template from .env.example):

OPENAI_API_KEY=your_api_key_here

Note: Make sure you have access to GPT-4 Turbo as the system uses gpt-4-turbo-preview model.

Running the Application

Run the main script from the project root directory:

python -m src.main

This will run two test cases:

1. Canceling an order: "I wish to cancel order_id 223"
2. Placing a new order: "customer_id: customer_14 : I wish to place order for item_51 with order quantity as 4 and domestic"

How It Works

The system uses a graph-based workflow with several key components:

1. Configuration

• Centralized configuration in config.py
• Environment variable management
• Shared LLM instance using GPT-4 Turbo

2. State Management

• Tracks order details, inventory status, shipping costs, and payment status
• Maintains conversation history and workflow progress

3. Workflow Nodes

• categorize_query: Determines user intent (place/cancel order)
• check_inventory: Verifies item availability
• compute_shipping: Calculates shipping costs based on location and weight
• process_payment: Simulates payment processing
• cancel_order: Handles order cancellation requests

4. Conditional Logic

• Routes requests to appropriate handlers based on user intent
• Manages workflow branching for different scenarios

Sample Data

Inventory Data

The system includes sample inventory data with:

• Item IDs
• Stock levels
• Item weights
• Prices

Customer Data

Sample customer data includes:

• Customer IDs
• Locations (local/domestic/international)
• Contact information

Cleaning the Project

This script ensures a clean installation by removing all compiled Python files, build artifacts, and the virtual environment before creating a fresh setup.

# Make the script executable
chmod +x clean.sh

# Run the cleaning script
./clean.sh

Contributing

Feel free to submit issues, fork the repository, and create pull requests for any improvements.

License

Apache 2.0

Acknowledgments

This project uses:

• LangGraph for workflow orchestration
• LangChain for LLM integration
• OpenAI's GPT-4 Turbo for natural language understanding