E-commerce AI Assistant

An e-commerce AI assistant built with Langchain.rb and OpenAI. This demo articulates the idea that business logic will now also live in prompts. A lot of modern software development is stringing services (classes and APIs) together. This demo illustrate how AI can assist in executing business logic and orchestrating calls to various services.

Video tutorial: https://www.loom.com/share/83aa4fd8dccb492aad4ca95da40ed0b2

Diagram

Installation

1. git clone
2. bundle install
3. cp .env.example .env and fill it out with your values.
4. Run sendmail in a separate tab.

Running

1. Run setup_db.rb to set up the database:

ruby setup_db.rb

2. Load Ruby REPL session with everything loaded:

ruby main.rb

3. Paste it the following code:

llm = Langchain::LLM::OpenAI.new(
  api_key: ENV["OPENAI_API_KEY"],
  default_options: { chat_completion_model_name: "gpt-4-turbo" }
)

# INSTRUCTIONS 1
new_order_instructions = <<~INSTRUCTIONS
  You are an AI that runs an e-commerce store called “Nerds & Threads” that sells comfy nerdy t-shirts for software engineers that work from home.

  You have access to the shipping service, inventory service, order management, payment gateway, email service and customer management systems. You are responsible for processing orders.

  FOLLOW THESE EXACT PROCEDURES BELOW:

  New order step by step procedures:
  1. Create customer account if it doesn't exist
  2. Check inventory for items
  3. Calculate total amount
  4. Charge customer
  5. Create order
  6. Create shipping label
  7. Send an email notification to customer
INSTRUCTIONS

# INSTRUCTIONS 2
return_order_instructions = <<~INSTRUCTIONS
  You are an AI that runs an e-commerce store called “Nerds & Threads” that sells comfy nerdy t-shirts for software engineers that work from home.

  You have access to the shipping service, inventory service, order management, payment gateway, email service and customer management systems. You are responsible for handling returns.

  FOLLOW THESE EXACT PROCEDURES BELOW:

  Return step by step procedures:
  1. Lookup the order
  2. Calculate total amount
  3. Refund the payment
  4. Mark the order as refunded
INSTRUCTIONS

# Create the assistant
assistant = Langchain::Assistant.new(
  # Instructions for the assistant that will be passed to OpenAI as a "system" message
  instructions: new_order_instructions,
  llm: llm,
  tools: [
    Langchain::Tool::InventoryManagement.new,
    Langchain::Tool::ShippingService.new,
    Langchain::Tool::PaymentGateway.new,
    Langchain::Tool::OrderManagement.new,
    Langchain::Tool::CustomerManagement.new,
    Langchain::Tool::EmailService.new
  ],
  # Thread instance that will keep track and record Messages
  thread: Langchain::Thread.new
)

# REQUESTS:

# Submit an individual order:
assistant.add_message_and_run content: "Andrei Bondarev (andrei@sourcelabs.io) just purchased 5 t-shirts (Y3048509). His address is 667 Madison Avenue, New York, NY 10065", auto_tool_execution: true

# Clear the thread
assistant.clear_thread!
# Reset the instructions
assistant.instructions = new_order_instructions

# Submit another order:
assistant.add_message_and_run content: """
New Order
Customer: Stephen Margheim (stephen.margheim@gmail.com)
Items: B9384509 x 2, X3048509 x 1
3 Leuthingerweg, Spandau, Berlin, 13591, Deutschland
""", auto_tool_execution: true

# Clear the thread
assistant.clear_thread!
# Set the new instructions
assistant.instructions = return_order_instructions

# Process a return:
assistant.add_message_and_run content: "stephen.margheim@gmail.com is returning order ID: 2", auto_tool_execution: true

# Clear the thread
assistant.clear_thread!
# Set the new instructions
assistant.instructions = return_order_instructions

# Updating inventory:
assistant.add_message_and_run content: """
INVENTORY UPDATE:
B9384509: 100 - $30
X3048509: 200 - $25
A3045809: 10 - $35
""", auto_tool_execution: true