Skip to Content
View all events

Building Multimodal Agents with Gemini, Google ADK, and NVIDIA Agent Toolkit Bootcamp

Published by O'Reilly Media, Inc.

Intermediate content levelIntermediate

Multimodal, real-time, and secure agents for production

What you’ll learn and how you can apply it

  • Architect multi-agent crews using parallel processing and single-responsibility patterns
  • Implement long-term memory and semantic search using knowledge graphs (Spanner)
  • Construct real-time voice/video agents using bidirectional streaming
  • Secure agents against prompt injection and data exfiltration using Model Armor and IAM

Course description

You already understand basic agent concepts. In this two-day bootcamp, you’ll turn them into action: building and deploying a secure, event-driven multi-agent system. Instructors Ayo Adedeji and Annie Wang replace abstract theory with a “Way Back Home” mission, where every coding exercise brings you one step closer to saving a crew stranded in space.

Using Google’s Agent Development Kit (ADK) and the latest Gemini models, you’ll build a complete agentic ecosystem. Day 1 focuses on foundations, orchestration, and memory (GraphRAG). Day 2 advances to real-time bidirectional streaming, event-driven architectures, and enterprise security patterns. At the end of this mission, you’ll have built and deployed a production-grade multi-agent system on Google Cloud.

This live event is for you because...

  • You’re a developer or AI engineer.
  • You work with LLMs and want to move from simple chatbots to complex, stateful agentic workflows.
  • You want to learn enterprise patterns for security, scalability, and multimodal interaction.

Prerequisites

  • A Google account (personal Gmail recommended for credit redemption)
  • A Google Cloud account (credits provided via Cloud Shell)
  • Access to a Chrome browser
  • A GitHub account (optional, for forking the repository)
  • Proficiency in Python
  • A basic understanding of LLMs (prompts, APIs)

Recommended follow-up:

Schedule

The time frames are only estimates and may vary according to how the class is progressing.

Day 1: Foundations and the Rescue

Mission briefing and identity (50 minutes)

  • Presentation: The agent landscape and ADK—moving beyond chatbots
  • Hands-on exercise: Build an avatar generator using Gemini to establish your explorer identity
  • Q&A
  • Break

Multi-agent orchestration (75 minutes)

  • Presentation: The orchestrator pattern—parallel versus sequential agents
  • Group discussion: When is a single agent sufficient and when do you need a crew?
  • Hands-on exercise: Build a specialist crew (geologist, botanist, astronomer) to triangulate location using parallel analysis
  • Q&A
  • Break

Knowledge graphs and RAG (60 minutes)

  • Presentation: Beyond vector search—introduction to GraphRAG with Spanner
  • Group discussion: Why do standard RAG approaches fail with complex, multihop queries?
  • Hands-on exercise: Ingest survivor data and build a semantic search engine using Spanner Graph
  • Break

Long-term memory (55 minutes)

  • Presentation: State management and memory banks in Vertex AI
  • Hands-on exercise: Connect the GraphRAG agent to long-term memory to personalize the rescue mission
  • Q&A

Day 2: Real-Time Systems and Security

Real-time multimodality (60 minutes)

  • Presentation: Bidirectional streaming—handling audio/video latency
  • Hands-on exercise: Build a real-time agent that sees and hears via webcam
  • Q&A
  • Break

Agent interoperability and deployment (60 minutes)

  • Presentation: Agent2Agent (A2A) protocols and discovery
  • Group discussion: Standardization in multi-agent systems
  • Hands-on exercise: Build a dispatcher agent that discovers and delegates tasks to a remote architect agent
  • Break

Event-driven architectures (60 minutes)

  • Presentation: Decoupling agents and server-sent events (SSE)
  • Hands-on exercise: Build a satellite coordination system using an asynchronous message bus
  • Q&A
  • Break

Agentic security and defense (55 minutes)

  • Presentation: Prompt injection, data exfiltration, and defense in depth
  • Hands-on exercise: Implement Model Armor guards and agent identity to harden the system against attacks

Wrap-up and Q&A (5 minutes)

Your Instructors

  • Ayo Adedeji

    Ayo Adedeji is a Developer Relations Manager at NVIDIA, specializing in bridging advanced AI technologies with practical developer solutions. With a background as an ML engineer in healthcare, Ayo’s expertise spans computational biology, big data processing, and foundation models. He holds engineering degrees from Stanford and Johns Hopkins and is passionate about helping developers across industries adopt accelerated computing to build innovative, responsible AI solutions.

  • Annie Wang

    Annie Wang is a software engineer, developer advocate, and internal career coach at Google, where she combines deep technical expertise with a passion for empowering developers. A sought-after speaker, she specializes in demystifying complex AI concepts, focusing on multi-agent systems, Google’s Agent Development Kit, and agent evaluation. She has delivered technical keynotes at major industry events including DevFest (NYC, Boston, Beijing) and moderates intensive AI courses with Kaggle. Annie is known for her “learn by doing” philosophy, helping engineers transition from theory to building practical, production-ready AI applications.

Skills covered

  • AI Agents
  • Cloud Native