LLM-Powered Email Classification and Order Processing

Overview

This project is a proof-of-concept application that leverages Large Language Models (LLMs) to classify customer emails and process orders, improving operational efficiency for a fashion retail store. The system automates email categorization, order fulfillment, and customer responses, reducing manual effort and enhancing the customer experience.

Objective

The goal is to build an AI-powered solution that intelligently processes email order requests and customer inquiries. The system will:

• Categorize emails into product inquiries or order requests.
• Process orders based on stock availability.
• Generate appropriate responses for each scenario.

Data Sources

The system interacts with a Google Spreadsheet containing:

• Product Catalog: A list of products with details like product ID, name, category, stock availability, description, and season.
• Customer Emails: A dataset of incoming emails with unique email IDs, subject lines, and message bodies.

Key Features

1. Email Classification

• Uses LLMs to classify each email as either a product inquiry or an order request.
• Outputs are saved in the email-classification sheet.

Example Output:

Email ID	Category
101	Product Inquiry
102	Order Request

2. Order Processing

• Verifies product availability for order requests.
• If in stock, the order is created and inventory is updated.
• If out of stock, the status is marked as out of stock.
• Results are stored in the order-status sheet.

Example Output:

Email ID	Product ID	Quantity	Status
102	P001	2	Created
103	P005	1	Out of Stock

3. Order Response Generation

• Generates email responses based on the order processing results.
• If an order is fulfilled, a confirmation email is sent.
• If an order is out of stock, an alternative solution is suggested.
• Responses are stored in the order-response sheet.

Example Output:

Email ID	Response
102	Your order for 2 units of P001 has been processed successfully.
103	Unfortunately, P005 is out of stock. Please consider an alternative.

4. Handling Product Inquiries

• Uses LLMs to answer customer queries based on the product catalog.
• Ensures scalability to handle large datasets without exceeding token limits.
• Stores responses in the inquiry-response sheet.

Example Output:

Email ID	Response
101	The P002 is a summer collection dress available in multiple sizes.

Implementation Details

• Uses GPT-4o via OpenAI API for email processing and response generation.
• Employs LangChain and other Python libraries for optimized AI-driven workflows.
• Handles large datasets efficiently by minimizing unnecessary API calls.

Solution Architecture Diagram

This diagram illustrates the end-to-end workflow for the LLM-powered automated email classification, order processing, and customer inquiry response system for a fashion store:

• Customer Emails are fetched and processed in sequence.
• Emails are classified using GPT-4o into:
  • Product Inquiries: answered using LLM-based retrieval and response generation.
  • Order Requests:
    • Order detail extraction using LLMs, grounded by product catalog data via LangChain SQL agents.
    • Order processing and stock validation, ensuring orders are fulfilled fully or partially while updating inventory.
    • LLM-based generation of clear, production-ready customer responses reflecting order status.

This architecture ensures scalable, intelligent, and automated handling of customer interactions while optimizing API token usage and aligning with enterprise-level prompt engineering best practices.

Technologies Used

• GPT-4o (via OpenAI API)
• LangChain
• Python (pandas, SQLAlchemy)
• Jupyter Notebook