Dual-Agent Sales Call Simulation

This project simulates a sales call between an AI sales agent (Sara) and a customer (John) who abandoned a shirt in their shopping cart. The AI interacts with the customer to complete the purchase, and a supervisor agent evaluates the conversation at the end to determine the next steps.

Features

• AI-powered sales agent (Sara) that engages with the customer.
• Customer intent detection using a language model.
• Supervisor agent that analyzes the conversation after it ends.
• Automatic handling of objections:
  • Price concerns → 10% discount.
  • Shipping concerns → Free shipping.
  • Quality concerns → Free returns.
  • Completed purchase → Send cart link.
• Conversation timeout after 5 minutes.

Installation

Prerequisites

• Python 3.x
• requests library
• Local instance of an LLM API (e.g., DeepSeek) running at http://localhost:1234

Setup

1. Clone this repository:

   git clone https://github.com/asantosca/dual-agent-sales.git
   cd dual-agent-sales
   

2. Install dependencies:

   pip install requests
   

3. Run the script:

   python abandoned_cart.py
   

How It Works

1. The sales agent (Sara) calls the customer and engages in a natural conversation.
2. Sara responds to objections and tries to close the sale.
3. If the customer says "goodbye" or 5 minutes pass, the conversation ends.
4. The supervisor agent analyzes the conversation and determines:
   • Whether to send an email with the cart link.
   • Whether a discount, free shipping, or free returns were applied.
   • Whether the customer abandoned the cart.
5. The final action is printed to the console.

Example Interaction

Hi John, this is Sara from ShirtsAndBelts. I noticed you left a blue medium shirt in your cart. Is there anything I can help you with?
John: It's too expensive.
Sara: I understand! I can offer you a 10% discount. Would that help?
John: Yes, that works.
Sara: Great! I’ll send a link to your email to complete the purchase.
Supervisor: Sending email with cart link.