GCP Agent Platform in 6 hours - Gen AI, Gemini, Agentic AI

Explore Google's generative ai landscape, including Gemini models, image and video tools like Imagine and Vue, and multi-modal capabilities, with model IDs, token limits, and lifecycle in Vertex AI.

Explore Vertex AI Model Garden to browse Gemini 3, Gemma, and other models, filter by type, view technical details, and deploy open models to Vertex AI or GKE.

Explore Vertex AI Studio as a developer-focused tool to build AI apps, choose models (Gemini or third-party), adjust thinking budget and safety settings, and generate code for Cloud Run.

Explore Vertex AI Studio in this walk-through, generating images, video, music, and code with Nano Banana Pro, and learn to use live API with Gemini to build AI apps.

Use Vertex AI Studio's build app to auto-generate an AI app with frontend and backend, download the code to run locally, and cartoonify uploaded photos with a Gen AI model.

Create and copy an api key in Vertex AI Studio to enable code use in your local development environment, then paste the key into your program for the next lecture.

Learn how to access a Gemini model from a Python program in a local environment by installing the Google Gen AI library, setting the API key, and streaming results.

Create and configure a service account to access a Gen AI model without an api key, granting Vertex AI user permissions and using a key for authentication.

Use your Google ID to authenticate to GCP with gcloud auth application-default login for local Vertex AI testing, avoiding keys; for production, consider service accounts and other auth options.

Compare Colab Enterprise and Workbench as Vertex AI notebooks offerings, explain why two options exist, and highlight lightweight Colab Enterprise versus heavier, feature-rich Workbench with Git and GCS integration.

Explore Colab Enterprise within Vertex AI, create a notebook in minutes, run programs without keys, and authenticate with your logged-in GCP ID to access models based on permissions.

Explore how to create and manage Colab Enterprise runtimes using templates, selecting machine types, idle time, and networking, then connect, run code, and schedule recurring executions.

Learn to create a Workbench instance, attach a notebook, and configure runtimes with machine types, GPUs, JupyterLab versions, and a service account for shared access.

Explore how Google Vertex AI Workbench offers JupyterLab with built-in ML libraries, Gemini CLI, git cloning, and seamless BigQuery integration, plus governance of service accounts for Vertex AI access.

Explore the workbench, noting that managed and user managed notebooks will be deprecated. Learn to create and view executions and schedules to run daily model inference from a notebook environment.

Learn to use Vertex AI and GCP services for end-to-end machine learning—data storage, training with AutoML or custom containers, and deployment via Vertex AI Endpoint, Cloud Run, or GKE.

Compare GCP machine learning products across storage, databases, and deployment options. Explore Vertex AI deployment options with Cloud Run, GKE, and Compute Engine.

Use Vertex AI to train, develop, and deploy models. Link data to a GCS bucket or BigQuery, train with AutoML or custom training, and deploy via endpoints from model registry.

Learn to create and configure datasets in Vertex AI, choosing region, data type, and training objective for image and tabular data, while recognizing text and video datasets are deprecated.

Explore regression training with AutoML by building a house price model from a CSV dataset, uploading to a GCS bucket, and creating a dataset for training and testing.

Create a tabular regression dataset in Vertex AI by importing a csv from a storage bucket or BigQuery, then review statistics and prepare for model training.

Train a regression model using AutoML, selecting the dataset and target column (house price), then set training options, budgets, data splits, and optional column exclusions.

Deploy an AutoML-trained tabular model to a Vertex AI endpoint, review the model registry and deployment options, and understand scaling and monitoring implications.

Test your AutoML model by invoking the REST endpoint with a sample JSON input and curl, or use the Python option, and stop or delete the endpoint to manage costs.

Discover custom training, where you control code, hyperparameters, and containers, moving beyond AutoML, with pre-built containers for TensorFlow, PyTorch, Scikit-learn, and XGBoost or a container; Vertex AI handles scaling.

Perform custom training with a pre-built container using tensorflow on a 10,000-row sentiment CSV from kegger.com, uploaded to a GCS bucket, to build a classifier that predicts sentiment levels.

Understand the project structure for custom training on Vertex AI, featuring a sentiment analysis package, trainer script, and packaging from a GCS bucket to save the model.

Train a sentiment analysis model with custom training in Vertex AI using TensorFlow 2.11, a custom container, and trainer.task as the entry point.

Complete custom training quickly to produce a TensorFlow sentiment model, save it to a bucket, import artifacts into a pre-built TensorFlow container, and prepare for deployment with TensorFlow 2.11.

Import a custom-trained model into model registry, then deploy it to sentiment analysis endpoint with 100% traffic on a single n1 standard 2 node. Use containers for training and deployment.

Test a deployed endpoint by sending a sample request, inspect sentiment predictions (negative, neutral, positive), and map results to levels while validating a rebuilt container serving a replica model.

Demonstrate vector space and embedding to show how LLM learns word relationships through tokenization and dimensional mappings, connecting dog, cat, king, and queen.

Compare vector search, rag engine, and Vertex AI search to show how index search differs from content search and llm-assisted summarization.

Explore Vertex AI vector search by creating and deploying an index endpoint, embedding data from a bucket with text embedding 005, and preparing 600 books for indexing.

Create a 768-dimension embedding index with cosine distance, set an index ID, choose batch or stream updates and an approximate neighbor count, then deploy a public endpoint in London.

Deploy and update your index with embeddings, then query the deployed endpoint to perform semantic vector search, retrieving similar items based on embeddings rather than keywords.

Explore how retrieval augmented generation grounds an llm with private data by converting it into vector space embeddings for semantic search, enabling llm-powered private data insights.

Configure a RegEngine in basic tier, create a corpus from Google Cloud Storage, attach two text files about global warming, and generate embeddings for vector search.

Ground the LLM with your own data in Vertex AI Studio using the RAC engine, query the corpus, and view grounded sources and confidence scores.

Explore Vertex AI Search as a higher-level layer over RegEngine, offering ready-made apps for custom, site, media, and commerce searches, including domain-specific options like healthcare or retail.

Create two datastores to ground the llm with your data using Vertex AI Search, crawling a public website like en.wikipedia.org and a cloud storage bucket.

Create and configure a custom search app with Vertex AI Search, linking multiple datastores, testing via curl, and integrating with websites through a widget or application programming interface.

Implement Gemini Enterprise (formerly Agentspace) by grounding private data with connected data stores, leveraging Gemini’s LLM and BucketStore or Wikistore to deliver grounded, data-driven answers.

discover how ai agents, powered by llms and rag with external APIs, plan trips by checking weather, comparing flights and hotels, and automatically book the best options.

Explore google adk as an open-source, cloud-integrated framework for building ai agents. It offers multimodal outputs, memory handling, multistep reasoning, and tool integrations within a multi-agent system.

Explore the GCP Vertex AI console options for ADK agents, focusing on four direct development paths: agent designer, agent garden, agent engine, and advanced MCP integration.

Learn to set up an ai agent project with the adk: create a folder, choose Gemini 2.5 plus, and configure a gemini api key or vertex ai backend.

Learn to authenticate the agent using Gemini API key or Google API key, create and manage API keys in a project, and set up the .env for Gemini integration.

Test and run your agent locally with ADK web, ADK run, and the API server, using the development UI, traces, and session management for debugging.

Learn how to set Vertex AI as the backend for agents, configure project id and region, manage authentication, and test multiple agents via adk web or adk run.

Deploy your agent to Vertex AI agent engine or other container environments with the ADK deploy command, create a history agent, and test with ADK run before deployment.

Test your deployed Vertex AI agent via the agent engine and playground to verify responses. Authenticate with IAM, create a session, and use curl to query the REST URL.

Explore the agent designer in Vertex AI and learn to create a UI-based agent, edit its description and instruction, and generate agent.py code from the preview.

Explore the agent garden in Vertex AI, filter for specific agents like a pdf parsing agent, and deploy them via cloud shell using the agent starter pack.

Use tools to empower llm-based agents with live data access, database queries, weather, and external services, leveraging built-in ADK tools like Google Search and custom functions.

A multi-agent system coordinates a root agent with sub-agents for hotel, flight, and itinerary, and uses A2A protocols to connect to external services.

Build a multi-agent system with a root agent coordinating SongAgent, StoryAgent, and JokeAgent. Describe sub-agents and modularize into directories; agent-to-agent configuration is not covered.

Learn to create, run, and deploy AI agents with Google ADK, explore multi-agent systems, memory, callbacks, and deployment options like agent engine, Cloud Run, and GKE.