AI Agents Bootcamp with ADK, GCP : Safety & Guardrails(2026)

Set up a clean, reproducible development environment for AI apps by creating and activating a Python virtual environment, installing NumPy, pandas, and Matplotlib, and using a requirements.txt and Jupyter kernel.

Set up your Google Cloud Platform account from scratch to access $300 free credits for 90 days and explore Vertex AI, storage, and other cloud services.

Build your first AI agent by installing Google Generative AI, configuring the API key from Google AI Studio, and running a loop with Gemini 2.5 to generate responses.

Build ai agents by combining llms, tools, and memory with guardrails and state management; follow the interface, orchestrator, reasoning, tools, memory, feedback, monitor, and output to create scalable systems.

Navigate the agent life cycle from safe deployment to continuous learning, employing staged rollouts, kill switches, trace correlation, service level objectives, drift monitoring, and evidence-based iteration.

Explore the agent execution loop—perception, understanding, planning, acting, observing, reflecting, and deciding—alongside debugging and observability practices to ensure reliable, safe ai operation.

Build your task agent from scratch by configuring a .env file for Gemini API key, loading variables, and using the 2.5 flash model with a one-sentence persona for fast responses.

Explore how agent tools convert thinking into acting for AI, and how the model decides when to use API, file, database, and system tools to connect AI to outside world.

Learn to build simple, single-responsibility tools that bridge models with external data, test them independently, structure inputs clearly, and prepare to connect tools to the model in the next lecture.

Discover how ai agents decide to call tools, map tools to questions, and execute calculations, revealing the core loop of tool-driven reasoning.

Learn to extend AI agents with file processing tools that read local files, extract context such as revenue data from sample.txt, and test file-based workflows.

Explore why memory matters for ai agents by moving from stateless conversations to memory-enabled interactions, covering short-term, long-term, and context memory.

Initialize a chat history to enable short-term memory, storing user queries and model responses; memory is temporary, with long-term memory using vector databases discussed later.

Discover context management strategies for AI agents by trimming chat history to a fixed max, using selective memory retrieval, and summarizing conversations to keep context focused and cost-efficient.

Build a context-aware agent by managing memory, trimming and summarizing chat history, and integrating tools to query weather and perform additions, demonstrating safe, scalable system design.

Implement safety filters as a protective layer between the model and its output, blocking unsafe content by categorizing hate speech, harassment, sexually explicit, and dangerous content.

Explore prompt injection attacks, where input manipulation overrides system behavior, and learn guardrails and input validation to prevent direct, indirect, tool, and data exfiltration attacks, including memory leakage concerns.

Explore why evaluating AI agents is hard, highlighting probabilistic outputs, context dependence, and multiple valid answers. Learn Rubik-based scoring to structure robust, multidimensional evaluation.

Learn how to design robust test cases for AI agents, covering prompts, categories, expected outcomes, memory, safety, tool usage, and attack resistance to ensure reliable performance.

Automate evaluation workflows for AI systems, test changes automatically, store metrics, compare with prior runs, and alert when performance drops, ensuring deployment only after validation.

Explore techniques to improve agent performance within AI agents bootcamp, focusing on safety and guardrails using ADK and GCP.

Master dynamic delegation through an orchestrator that assigns tasks to the right agent at the right time with context. Learn how user intent, capabilities, and safety guide decisions.

Identify planner and orchestrator as the core components of multi-agent systems, where the planner creates a structured plan and the orchestrator executes it by routing tasks to agents.

Boost developer productivity with the Gemini Code Assist productivity assistant within an AI agents bootcamp that emphasizes safety and guardrails for ADK and GCP.

Explore voice transcription and call insight analysis to support safe, compliant AI agents with built-in guardrails. Understand how transcription data turns into actionable insights for monitoring and controlling AI behavior.

Build a multimodel intelligent agent using Vertex AI that searches, analyzes, and generates structured answers with tools and external data for real-world tasks.

Master observability to monitor AI systems, debug step by step with logs, metrics, and traces, and manage stateless autoscaled deployments with request IDs.

Secure ai systems rely on virtual private clouds to isolate services, enforce access control, keep databases private, and apply zero-trust architectures with iam, jwt, mtls, and iap.

Contrast demo and production AI systems, highlighting reliability, scalability, load balancing, and authentication, rate limiting, and data validation in an end-to-end, stateless, async architecture.