This project implements an AI-powered customer service agent for Cymbal Home & Garden, a big-box retailer specializing in home improvement, gardening, and related supplies. The agent is designed to provide excellent customer service, assist customers with product selection, manage orders, schedule services, and offer personalized recommendations.

The Cymbal Home & Garden Customer Service Agent is designed to provide a seamless and personalized shopping experience for customers. It leverages Gemini to understand customer needs, offer tailored product recommendations, manage orders, and schedule services. The agent is designed to be friendly, empathetic, and highly efficient, ensuring that customers receive the best possible service.

The key features of the Customer Service Agent include:

The agent is built using a multi-modal architecture, combining text and video inputs to provide a rich and interactive experience. It mocks interactions with various tools and services, including a product catalog, inventory management, order processing, and appointment scheduling systems. The agent also utilizes a session management system to maintain context across interactions and personalize the customer experience.

It is important to notice that this agent is not integrated to an actual backend and the behaviour is based on mocked tools. If you would like to implement this agent with actual backend integration you will need to edit customer_service/tools.py

Because the tools are mocked you might notice that some requested changes will not be applied. For instance newly added item to cart will not show if later a user asks the agent to list all items.

• Personalized Customer Assistance:
  • Greets returning customers by name and acknowledges their purchase history.
  • Maintains a friendly, empathetic, and helpful tone.
• Product Identification and Recommendation:
  • Assists customers in identifying plants, even from vague descriptions.
  • Requests and utilizes visual aids (video) to accurately identify plants.
  • Provides tailored product recommendations based on identified plants, customer needs, and location (e.g., Las Vegas, NV).
  • Offers alternatives to items in the customer's cart if better options exist.
• Order Management:
  • Accesses and displays the contents of a customer's shopping cart.
  • Modifies the cart by adding and removing items based on recommendations and customer approval.
  • Informs customers about relevant sales and promotions.
• Upselling and Service Promotion:
  • Suggests relevant services, such as professional planting services.
  • Handles inquiries about pricing and discounts, including competitor offers.
  • Requests manager approval for discounts when necessary.
• Appointment Scheduling:
  • Schedules appointments for planting services (or other services).
  • Checks available time slots and presents them to the customer.
  • Confirms appointment details and sends a confirmation/calendar invite.
• Customer Support and Engagement:
  • Sends via sms or email plant care instructions relevant to the customer's purchases and location.
  • Offers a discount QR code for future in-store purchases to loyal customers.
• Tool-Based Interactions:
  • The agent interacts with the user using a set of tools.
  • The agent can use multiple tools in a single interaction.
  • The agent can use the tools to get information and to modify the user's transaction state.
• Evaluation:
  • The agent can be evaluated using a set of test cases.
  • The evaluation is based on the agent's ability to use the tools and to respond to the user's requests.

The agent's session state is preloaded with sample customer data, simulating a real conversation. Ideally, this state should be loaded from a CRM system at the start of the conversation, using the user's information. This assumes that either the agent authenticates the user or the user is already logged in. If this behavior is expected to be modified edit the get_customer(current_customer_id: str) in customer.py

The agent has access to the following tools:

• send_call_companion_link(phone_number: str) -> str: Sends a link for video connection.
• approve_discount(type: str, value: float, reason: str) -> str: Approves a discount (within pre-defined limits).
• sync_ask_for_approval(type: str, value: float, reason: str) -> str: Requests discount approval from a manager.
• update_salesforce_crm(customer_id: str, details: str) -> dict: Updates customer records in Salesforce.
• access_cart_information(customer_id: str) -> dict: Retrieves the customer's cart contents.
• modify_cart(customer_id: str, items_to_add: list, items_to_remove: list) -> dict: Updates the customer's cart.
• get_product_recommendations(plant_type: str, customer_id: str) -> dict: Suggests suitable products.
• check_product_availability(product_id: str, store_id: str) -> dict: Checks product stock.
• schedule_planting_service(customer_id: str, date: str, time_range: str, details: str) -> dict: Books a planting service appointment.
• get_available_planting_times(date: str) -> list: Retrieves available time slots.
• send_care_instructions(customer_id: str, plant_type: str, delivery_method: str) -> dict: Sends plant care information.
• generate_qr_code(customer_id: str, discount_value: float, discount_type: str, expiration_days: int) -> dict: Creates a discount QR code.

• Python 3.10+

• uv for dependency management and packaging

  • See the official uv website for installation.

  curl -LsSf https://astral.sh/uv/install.sh | sh

Use the Agent Starter Pack to scaffold a production-ready project and choose your deployment target (Vertex AI Agent Engine or Cloud Run), with CI/CD and other production features. The easiest way is with uv (one command, no venv or pip install needed):

uvx agent-starter-pack create my-customer-service -a adk@customer-service

If you don't have uv yet: curl -LsSf https://astral.sh/uv/install.sh | sh

The starter pack will prompt you to select deployment options and set up your Google Cloud project.

# Create and activate a virtual environment
python -m venv .venv && source .venv/bin/activate # On Windows: .venv\Scripts\activate

# Install the starter pack and create your project
pip install --upgrade agent-starter-pack
agent-starter-pack create my-customer-service -a adk@customer-service

From your newly created project directory (e.g. my-customer-service), run:

cd my-customer-service
uv sync --dev
uv run adk run customer_service

For the web UI:

uv run adk web

Then select customer_service from the dropdown menu.

uv sync --dev
uv run pytest eval

uv run pytest tests/unit

import vertexai
from customer_service.config import Config
from vertexai.preview.reasoning_engines import AdkApp


configs = Config()

vertexai.init(
    project="<GOOGLE_CLOUD_LOCATION_PROJECT_ID>",
    location="<GOOGLE_CLOUD_LOCATION>"
)

# get the agent based on resource id
agent_engine = vertexai.agent_engines.get('DEPLOYMENT_RESOURCE_NAME') # looks like this projects/PROJECT_ID/locations/LOCATION/reasoningEngines/REASONING_ENGINE_ID

for event in remote_agent.stream_query(
    user_id=USER_ID,
    session_id=session["id"],
    message="Hello!",
):
    print(event)

Here's a quick example of how a user might interact with the agent:

User:: hi there

Agent: Hey Alex, I see you've been a Cymbal Home & Garden customer for a couple of years now - we appreciate your business! I also see you have a few things in your cart already. How can I help you get your garden growing today?

User: Yeah, so I'm a bit confused. I just bought some...uh...plants or flowers that I wanted to put in the backyard. They just say "sun loving annuals" on them. I totally spaced on picking up potting soil so I'm placing an order for pickup now. The problem is, I'm not sure if the potting soil and fertilizer I picked out are the right ones.

Agent: I understand – it can be tricky when you're not quite sure what you've got! Let's figure this out together. Don't worry, we'll get your plants the right care. To best help you, would you be willing to share a quick video of the plants you've purchased? That will help me identify them and make sure you have the right supplies.

User: Sure, how do I do that?

Agent: I just texted you a link to the number in your profile. Just click on that, and it'll open a secure video connection.

print(send_call_companion_link(phone_number='+12065550123'))

{'status': 'success', 'message': 'Link sent to +12065550123'}

User: Okay, I clicked the link...can you see this?

Agent: Yes I can see you Alex! Can you go ahead and point the camera at your plants so I can inspect them?

User: Yes of course! ...

This agent sample is provided for illustrative purposes only and is not intended for production use. It serves as a basic example of an agent and a foundational starting point for individuals or teams to develop their own agents.

This sample has not been rigorously tested, may contain bugs or limitations, and does not include features or optimizations typically required for a production environment (e.g., robust error handling, security measures, scalability, performance considerations, comprehensive logging, or advanced configuration options).

Users are solely responsible for any further development, testing, security hardening, and deployment of agents based on this sample. We recommend thorough review, testing, and the implementation of appropriate safeguards before using any derived agent in a live or critical system.