Gemma 4 production stack: Model Armor, ADK Agents, Tracing

Google Cloud Tech · Intermediate ·🤖 AI Agents & Automation ·2w ago
GCP credit → https://goo.gle/handson-ep8-lab1 Lab → https://goo.gle/guardians Gemma 4 is deployed. Now we secure it, build an agent on it, and make it observable. 🛡️ Model Armor — scans every prompt and response for jailbreaks, PII, and harmful content via a Load Balancer Service Extension. 🤖 ADK Agent — built with Agent Development Kit, powered by vLLM, deployed to Cloud Run via CI/CD. 📊 Prometheus Sidecar — scrapes vLLM metrics: token throughput, GPU utilization, latency. 🔍 Cloud Trace — OpenTelemetry tracing through the agent, end to end. Security. Agents. Observability. One stack. 👇 Chapters: 0:00 - Intro 8:25 - Erecting the shield of SecOps: Setup Model Armor 38:37 - Raising the watchtower: Agent pipeline 57:17 - The palantir of performance: Metrics and tracing 1:07:46 - The boss fight 1:12:30 - Wrap up More resources: Agent Development Kit (ADK) docs → https://goo.gle/4uflScr Model Armor documentation → https://goo.gle/4mz57Ga Cloud Trace documentation → https://goo.gle/4euYyCB Watch more Hands on AI → https://www.youtube.com/watch?v=qCBreTfjFHQ&list=PLIivdWyY5sqKnJOvP89yF8t9mWuzMTcbM 🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech #Gemma4 #ModelArmor Speakers: Ayo Adedeji, Annie Wang Products Mentioned: Cloud Run, Cloud Build, Model Armor, Agent Development Kit, Cloud Trace, Gemma 4
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Related AI Lessons

Voice AI Has a Networking Problem Nobody Talks About
Voice AI has a hidden networking problem that affects its performance, and understanding this issue is crucial for developers
Medium · AI
Voice AI Has a Networking Problem Nobody Talks About
Voice AI has a hidden networking problem that hinders its performance, and it's time to address it
Medium · Programming
Before an Agent Pays Anyone, Someone Has to Approve It: A Builder’s Read on FluxA
Learn how to manage agent payments with approval workflows using FluxA, a crucial step in deploying autonomous agents
Dev.to AI
The Payment Rail Problem for AI Agents, and Why FluxA Splits Wallet Control from Spend Access
Learn how FluxA solves the payment rail problem for AI agents by splitting wallet control from spend access, ensuring safer and more efficient transactions
Dev.to AI

Chapters (6)

Intro
8:25 Erecting the shield of SecOps: Setup Model Armor
38:37 Raising the watchtower: Agent pipeline
57:17 The palantir of performance: Metrics and tracing
1:07:46 The boss fight
1:12:30 Wrap up
Up next
Why AI engineering needs old-school discipline
The New Stack
Watch →