多智能体AI应用开发课程

Develop Multi AI Agent Applications with Gemini Agent ADK

Design, build and evaluate multi-agent LLM applications with Google's Agent Development Kit and Gemini.

About the course

Courseware and hands-on lab repository for the WSQ course Develop Multi AI Agent Applications with Gemini Agent ADK, built on Google's open-source Agent Development Kit (ADK) and the Gemini model family.

Learning Outcomes

Course Topics

Quick Start

Prerequisites: Python 3.13+, uv, and a free Gemini API key from Google AI Studio.

git clone https://github.com/tertiarycourses/TGS-2024042961-Develop-Multi-AI-Agent-Applications-with-Gemini-Agent-ADK.git
cd TGS-2024042961-Develop-Multi-AI-Agent-Applications-with-Gemini-Agent-ADK/labs
uv sync

Create a .env file in the labs/ folder:

GOOGLE_GENAI_USE_VERTEXAI=0
GOOGLE_API_KEY=your-google-api-key
OPENWEATHER_API_KEY=your-openweather-key   # optional, tool labs
TAVILY_API_KEY=your-tavily-key             # optional, search labs

Never commit your .env file or API keys. It is git-ignored in this repository.

Run any agent:

uv run adk run <agent_folder>   # terminal chat
uv run adk web                  # browser IDE at http://localhost:8000

Labs

Every lab is a self-contained folder holding its own agent script, data files and a README.md lab sheet. See labs/LABS.md for the full index.

Core ADK Patterns

Define an agent

from google.adk.agents import Agent

root_agent = Agent(
    model='gemini-2.0-flash',
    name='root_agent',
    description='A helpful assistant for user questions.',
    instruction='Answer clearly and concisely.',
)

Add a tool — the docstring and type hints are the contract the model reads.

def get_weather(city: str) -> dict:
    """Retrieves the current weather for a specified city.

    Args:
        city (str): The name of the city.

    Returns:
        dict: status and result or error msg.
    """
    return {"status": "success", "report": "..."}

agent = Agent(..., tools=[get_weather])

Multi-agent handoff

root_agent = Agent(
    name='root_agent',
    sub_agents=[math_tutor_agent, physics_tutor_agent, history_tutor_agent],
    instruction='Route each question to the right specialist.',
)

Sequential workflow

from google.adk.agents import SequentialAgent

workflow = SequentialAgent(
    name='workflow_agent',
    sub_agents=[input_agent, research_agent, report_agent],
)

Guardrail — return None to allow, an LlmResponse to block.

def block_keyword_guardrail(callback_context, llm_request):
    if "BLOCK" in last_user_message.upper():
        return LlmResponse(content=types.Content(
            role="model", parts=[types.Part(text="I cannot process this request.")]))
    return None

agent = Agent(..., before_model_callback=block_keyword_guardrail)

Structured output — note an agent with output_schema cannot also use tools.

from pydantic import BaseModel

class Recipe(BaseModel):
    title: str
    ingredients: list[str]
    cooking_time: int

agent = Agent(..., output_schema=Recipe)

Courseware

The Learner Guide carries the full step-by-step instructions for all 18 labs, plus reference sections on core ADK patterns, evaluating a RAG pipeline, and assessing the feasibility of an agent application. The deck includes the "Login to LMS and Download Course Material" slide showing the current LMS-TMS sign-in page.

The Trainer Slides (v1.4, 163 slides) teach each lab as a four-part unit — briefing → process map → procedure with the actual commands → verification with troubleshooting — alongside comparison matrices, decision maps, worked code examples and native charts. Slide transitions are deliberately restrained (content fades, section dividers push), with click-through reveals on the process maps so a stage can be discussed before the next appears.

Public/private distribution boundary: the assessment set (question papers and answer keys), the Facilitator Guide, the Assessment Plan and trainer reference material are confidential and are not published in this repository. Answer keys are never linked from the LMS.

Resources

• Google ADK Documentation
• Google AI Studio — free Gemini API keys
• Course page
• LMS / TMS

Support

Tertiary Infotech Academy Pte Ltd · UEN 201200696W Email: enquiry@tertiaryinfotech.com · Tel: +65 6100 0613 · tertiarycourses.com.sg

© 2026 Tertiary Infotech Academy Pte Ltd. All rights reserved.