Multi-Agent Systems with Google ADK and Vertex AI

Get to know your course instructors, set up GCP resources, and get an overview of the course.

Explain the core components of multi-agent systems and how to design their high-level architecture.

Design multi-agent systems using the Orchestrator–Worker pattern and formalize agent capabilities with A2A Agent Cards to enable standardized service communication.

Develop a multi-agent system by coding the designed architecture and connecting agents with well-defined interfaces.

Implement multi-agent architectures with ADK by creating specialized sub-agents with focused tools, orchestrators that route intent-based tasks, and structured delegation workflows.

Apply orchestration techniques to coordinate multiple agent actions and achieve complex workflows.

Build complex business processes using SequentialAgent for ordered pipelines and ParallelAgent for concurrent execution, optimizing workflows through intelligent task dependency management.

Configure routing mechanisms to manage data flow among agents in multi-agent systems.

Implement deterministic business rule routing with CustomAgent to inspect events and programmatically enforce compliance, discounts, and strict logic without LLM uncertainty.

Evaluate methods for tracking and updating agent state across multi-turn interactions.

Master explicit session state management using ToolContext and InvocationContext to persist complex data across multi-turn conversations with cloud-based Vertex AI Agent Engine.

Develop a coordinated multi-agent system that synchronizes states for coherent task execution.

Build distributed multi-agent microservices using RemoteA2aAgent for A2A communication, Agent Cards for service discovery, and MCP Database Toolbox for shared MySQL state across independent services.

Extend RAG to multiple cooperating agents, each specialized in certain retrieval tasks.

Implement enterprise multi-agent RAG systems that combine Vertex AI Search for unstructured docs with SQL databases for structured data, enabling hybrid knowledge sharing across agents.