Model Online / On-Site
Duration
Modules 5 modules

Course Objective & Scope

Cramming a complex business process into a single monolithic prompt with dozens of tools degrades tool-selection accuracy and makes testing nearly impossible. This programme teaches how to build modular, testable, hierarchical multi-agent systems using Google's open-source Agent Development Kit (ADK).

Core ADK primitives—including LlmAgent alongside deterministic SequentialAgent, ParallelAgent and LoopAgent orchestrators—shared session.state and artifact management, function/OpenAPI/MCP tools, guardrail callbacks, interactive debugging in the built-in ADK Web UI and trajectory evaluation are covered hands-on. Participants build a working multi-agent application for an internal business workflow.

Who Should Attend?

  • AI and software engineers building modular multi-agent applications in Python or Java
  • Developers architecting enterprise assistants on Vertex AI and Google Cloud
  • Technical teams combining predictable workflow orchestration with dynamic LLM routing
  • Prerequisite: intermediate Python programming and basic LLM / tool calling familiarity

Key Takeaways

  • Compose hierarchical agent trees mixing LlmAgent reasoning with deterministic SequentialAgent, ParallelAgent and LoopAgent flows
  • Coordinate sub-agents via LLM-driven delegation and structured session.state sharing
  • Integrate Python functions, OpenAPI specs, MCP servers and wrapped sub-agents (AgentTool)
  • Enforce safety policies, caching and human confirmation using model and tool lifecycle callbacks
  • Debug event traces in adk web and run automated trajectory evaluations with adk eval

Curriculum

Google ADK Architecture and Core Agent Types
  • LlmAgent, deterministic Workflow Agents (Sequential, Parallel, Loop) and custom agents
  • Local development and event-stream inspection with adk web, adk run and adk api_server
The ADK Tool Ecosystem: Functions, OpenAPI, MCP and AgentTool
  • Using ToolContext, long-running async tools and generating toolsets from OpenAPI specs
  • Connecting external MCP servers via MCPToolset and wrapping specialist agents as AgentTools
Hierarchical Multi-Agent Design, State and Artifacts
  • Structuring sub_agents hierarchies and dynamic LLM-driven peer/child delegation
  • Managing Session, scoped State prefixes (temp/user/app), Memory and binary Artifacts
Lifecycle Callbacks, Guardrails and Bidirectional Streaming
  • Intercepting execution with before/after model and tool callbacks for policy checks and guardrails
  • Real-time bidirectional audio and video streaming with the Gemini Live API
Automated Evaluation (adk eval) and Production Deployment
  • Authoring evalsets to assert both tool-call trajectories and final response quality
  • Deploying ADK agents to Cloud Run, GKE or managed Vertex AI Agent Engine

Other Courses in This Category

  • Enterprise Agent Development with Google Cloud Vertex AI Agent Engine

    Learn hands-on how to design, deploy, and operate scalable, secure, and managed enterprise agent architectures using Google Cloud Vertex AI Agent Engine and Agent Development Kit (ADK).

    Enterprise AI Platforms & Agent Cloud Ecosystem Corporate
    Live Course
    4 modules
    View
  • Vertex AI Model Garden: Enterprise Model Selection and Endpoint Management

    Learn how to evaluate, customize, deploy as secure endpoints, and optimize cost-performance for Gemini, partner models, and open-weight models on Vertex AI Model Garden.

    Enterprise AI Platforms & Agent Cloud Ecosystem Corporate
    Live Course
    4 modules
    View
  • Governance and Observability in Cloud-Based Agent Systems

    Learn how to design and implement end-to-end tracing, telemetry, automated evaluation, safety guardrails, and IAM-based access governance in enterprise cloud agent systems.

    Enterprise AI Platforms & Agent Cloud Ecosystem Corporate
    Live Course
    4 modules
    View
  • Enterprise AI Lifecycle with the IBM watsonx Platform

    Learn how to build enterprise AI solutions, feed them with hybrid data architectures, and govern them for regulatory compliance using IBM watsonx.ai, watsonx.data, and watsonx.governance.

    Enterprise AI Platforms & Agent Cloud Ecosystem Corporate
    Live Course
    4 modules
    View
  • Enterprise Digital Workers and Automation with IBM watsonx Orchestrate

    Learn hands-on how to design and manage digital assistants, custom skill catalogs, and multi-step workflow automations integrated with enterprise applications using IBM watsonx Orchestrate.

    Enterprise AI Platforms & Agent Cloud Ecosystem Corporate
    Live Course
    4 modules
    View
  • Stateful Multi-Agent Architectures and Production Orchestration with LangGraph

    Build production-grade multi-agent systems with cyclic workflows, persistent state checkpointing, human-in-the-loop approval flows, and LangSmith observability using LangGraph.

    Enterprise AI Platforms & Agent Cloud Ecosystem Corporate
    Live Course
    4 modules
    View

Get in touch