Adk MCP Gemma3

🚀 YouTube Search Project: A Comprehensive Guide to Integrating ADK with Gemma3 Model

This project is a full-fledged implementation of YouTube search functionality, leveraging the ADK framework and the 12B Gemma3 language model. It demonstrates how to enable large language models (LLMs) to interact with external APIs (such as SERP API) through a function call mechanism, thereby accomplishing complex web search tasks.

✨ Features

Core Modules

• ADK Framework: Provides LLM agent interfaces and tool management capabilities.
• Gemma3 12B Model: Used for understanding and generating natural language instructions.
• SERP API: Enables structured searches for YouTube videos.
• Result Formatting: Organizes API-returned results into a user-friendly list format.

Functional Modules

Search Agent Functionality

class YouTubeSearchAgent:
    def __init__(self, model_name="gema3:12b"):
        self.llm = LlmAgent(model_name)
        self._setup_agent()

    def _setup_agent(self):
        # Initialize the LLM agent and set up the YouTube search tool
        self.llm.init()
        self.llm.add_tool(youtube_search_tool)

    async def search_videos(self, query: str):
        response = await self.llm.run(query)
        return parse_youtube_response(response)

Search Tool Definition

def youtube_search_function():
    @llm_tool
    async def search_videos(query: str) -> dict:
        """Search for videos via the YouTube API"""
        results = await serpapi_call(query)
        return format_youtube_results(results)
    
    return {
        "name": "youtube_search",
        "function": search_videos,
        "description": "Perform a YouTube video search and return structured results"
    }

Result Parsing

def parse_youtube_response(response: str) -> dict:
    # Parse the JSON-formatted response returned by the LLM
    try:
        return json.loads(response)
    except JSONDecodeError:
        raise ValueError("Invalid JSON response")

📚 Documentation

Project Architecture

project/
├── Images/
│   ├── adk-gemma3.gif           # Application demonstration animation
│   └── adk-gemma-web.gif        # Web interface showcase
├── search/
│   ├── __init__.py              # Package initialization file
│   └── agent.py                 # Main agent implementation module
├── .gitignore                   # Git ignore rules
├── README.md                    # Project documentation
├── LICENSE                      # MIT open-source license
└── requirements.txt             # Dependency management file

Typical Application Scenarios

• Video Content Retrieval: Quickly find YouTube videos on specific topics.
• Educational Aid: Assist learners in finding teaching video resources.
• Market Research Support: Provide video materials for corporate competitor analysis.
• Content Creation Reference: Offer inspiration and reference materials for creators.

Project Characteristics

• Modular Design: Each functional unit is independent and scalable.
• High Configurability: Supports integration with multiple APIs.
• Intelligent Search: Semantic understanding powered by LLM.
• Result Optimization: Continuously improves search quality based on a feedback mechanism.

📦 Installation

1. Install dependencies:

pip install -r requirements.txt

2. Initialize the model:

ollama init gema3:12b

💻 Usage Examples

Basic Usage

agent = YouTubeSearchAgent()
results = agent.search_videos("Learning Python programming")
print(results)

🔧 Technical Details

Troubleshooting

Model Loading Failure

• Issue: Unable to initialize the Gemma3 model.
• Solution: Check the network connection and ensure the model is correctly installed.

API Call Exception

• Issue: The SERP API returns an error.
• Solution: Confirm that the API key is valid and check the quota limit.

Result Parsing Failure

• Issue: The response format does not meet expectations.
• Solution: Ensure that the LLM output conforms to the JSON specification.

📄 License

This project is licensed under the MIT open-source license, allowing free use and modification while retaining the copyright notice.

📇 Author Information

• Project Maintainer: [Your Name]
• Contact Email: your.email@example.com
• GitHub Repository: [Link]

This documentation comprehensively introduces how to build a fully functional YouTube search system using the ADK framework and the Gemma3 model. Through a modular design and clear code structure, it showcases the powerful capabilities of LLMs in practical applications.