A hands-on workshop exploring AI agent patterns and implementations using Google's agent frameworks.

Explore fundamental agent patterns using Google ADK (Python) and Firebase Genkit (TypeScript):

• ADK Agents: Simple, Orchestrator, Sequential, Parallel, and Loop patterns
• Genkit Agents: Custom tools, multi-modal processing, self-evaluation, and workflows

Topics covered:

• Agent orchestration and coordination
• Tool integration (YouTube API, Google Search, image generation)
• State management and data flow
• Iterative refinement and feedback loops
• Multi-modal AI (video, text, images)

This session builds on the fundamentals by constructing a practical, multi-agent system for a virtual coffee shop.

• Barista Agent System: An advanced implementation featuring an orchestrator managing specialized agents:
  • Head Barista: Menu queries and availability checks via semantic search (MCP Server)
  • Creative Director: Image generation with Imagen 3 and promotions management
  • Market Analyst: Global trend analysis using BigQuery and Wikipedia pageview data (MCP Toolbox)
• Data Ingestion Pipeline: Processes menu data with Gemini embeddings and stores vectors in Firestore for RAG retrieval.

Topics covered:

• Complex multi-agent collaboration with role-based delegation
• System design for specialized agent roles
• MCP (Model Context Protocol) integration for remote tools
• MCP Toolbox for database access (BigQuery)
• RAG implementation with Firestore vector search
• Data ingestion pipelines with embeddings
• Real-time trend analysis and data-driven recommendations

Each session contains detailed READMEs with setup instructions and examples.

cd Session-1/ADK                          # Python agents with Google ADK
cd Session-1/genkit                       # TypeScript agents with Firebase Genkit
cd Session-2/barista-agent-system         # Multi-agent coffee shop system
cd Session-2/pipeline-data-ingestion-menu # Menu data pipeline with embeddings

• Understand different agent orchestration patterns
• Build agents with custom tools and APIs
• Implement multi-agent systems with role-based specialization
• Apply feedback loops for quality improvement
• Work with multi-modal AI capabilities
• Integrate Model Context Protocol (MCP) for remote tool access
• Build RAG systems with vector embeddings and Firestore
• Connect agents to external data sources (BigQuery, APIs)

• Python 3.10+ and/or Node.js 20+
• Google AI API keys (Gemini)
• Google Cloud Platform account (for Session 2)
  • Firestore database
  • Cloud Storage bucket
  • BigQuery access (public datasets)
• Poetry for Python dependency management
• Basic understanding of async programming

Start with Session 1 to explore agent patterns!