GCP Vertex AI | Google AI & ML | Agentic AI (ADK)| MCP | A2A

Explore options to access gen AI models via web console, code, and fine-tuning with Vertex AI and Google AI Studio, emphasizing lightweight prototyping for beginners.

Explore Google AI Studio's playground to prompt Gemini models, switch models, and tailor outputs with temperature, thinking mode, and system instructions.

Learn how to access models outside Google AI Studio using the Gemini API key, and generate code in your language of choice to prompt the model from your program.

Create an API key in Google AI Studio, and manage projects, billing, and usage. Choose between free tier and tier one projects, and link billing in the Google Cloud Console.

Create and name an API key in Google AI Studio, select a project and tier, view the full key id, and copy or delete it for use in Python.

Access a Gemini model from Python with the Google Gen AI SDK by installing the SDK, setting the API key in the environment, and streaming prompt responses.

Create your own ai apps in Google AI Studio using the build ai apps feature, generate ready-to-run code, and deploy Tuneify AI from an uploaded image.

deploy your AI app using Google AI Studio's deploy app, link it to Cloud Run, and share a live URL; then delete services to avoid costs.

Explore authenticating to Vertex AI with GCP beyond Gemini keys, using service accounts, Workload Identity, and notebooks like Colab Enterprise and AI workbench.

Authenticate to GCP with a service account from IAM to access a Gemini model in your local Python code, downloading a JSON key and applying minimal Vertex AI user permissions.

Learn how to authenticate to GCP as an individual user using gcloud, application default credentials, and IAM to access Vertex AI models from your laptop or cloud shell.

Learn to use the Google Gen AI SDK and Google JNI library to access Gemini and other models, authenticate with a Gemini API key, and generate content.

Explore how token counts and model choice affect costs in Gemini, including input, output, and candidate tokens, plus the pro thinking budget and its impact.

Learn to limit the model's thinking budget using the generate content config, disable or cap thinking with zero, 300, or higher values to manage tokens and cost.

Limit the language model's output tokens with max output tokens and understand how truncation shapes the final response, while considering the thinking budget.

Explore how to manage language model randomness by adjusting temperature and top P/top K within Google's model ranges, using the Generate Content config for math precision or creative tasks.

Learn to manage safety settings in the generate content config, setting thresholds for harm, hate speech, harassment, and other categories to block outputs.

Set up system instructions to assign a persona to a language model, such as a history expert, for detailed outputs. Use a generate content config class to share context.

Learn to implement streaming responses with generate content stream in Python, printing the language model's output word by word and enabling interactive, chart-like applications.

Explore nano banana, Google's Gemini 2.5 flash image model with an image extension, a fast multimodal tool that creates high-quality images, like a cat riding a bike.

Edit an image using the nano banana multimodal model with an image and a text prompt, then save the edited cat image with a new name showing a red shoe.

Generate a five-second video from a prompt using the Veo model, then download the mp4 to a local directory, and note token usage and cost for longer prompts.

Learn to summarize a YouTube video using Gemini 2.5 flash by feeding a URL or an uploaded file, and generate a concise three-sentence video summary.

Discover how grounding with external sources, especially Google Search, lets a language model access real-time data and overcome knowledge cutoffs.

Explore using a custom function tool with an llm to schedule meetings by defining a schedule meeting function with attendees, date, time, topic, and let the model generate invocation arguments.

Explore what open models are and how they differ from closed models, including open weight options you can download and deploy locally or in cloud, with privacy implications.

Use OLAMA to run open weight models locally, download LLAMA and JEMA models, and manage them with OLAMA commands for chat-like interaction and self-hosted versus cloud options.

Deploy an open model on Cloud Run by selecting the Gemma model image, configuring 4 GB memory and 2 CPUs, and testing access via the deployed URL.

See Vertex AI rebrand to the agent platform and the notebooks menu now listing Colab Enterprise and Workbench; notebook content remains unchanged.

Explore GCP notebook options: Colab Enterprise and Workbench. Choose Colab Enterprise for lightweight, quick-start training; opt for Workbench for a JupyterLab IDE with deeper training and GPU or TPU support.

Explore Colab Enterprise within Vertex AI, learn to create or import notebooks, access sample notebooks and the gallery, filter by Gemini, and review runtimes, templates, and execution schedules.

Explore how Colab Enterprise uses runtimes and runtime templates to create and manage virtual machines for notebooks, including default and custom templates, idle shutdown, and cost considerations.

Schedule and execute notebook tasks in Colab Enterprise by creating recurring or one-time intervals, using cron-like frequencies, and saving outputs to a storage bucket.

Learn how Colab Enterprise schedules and executes notebook runs, view outputs in cloud bucket paths and timestamps, and configure templates and intervals for scalable executions.

Connect a notebook to a Google Cloud Storage bucket, read a CSV with the storage client, and optionally write files, enabling easy training workflows from Colab Enterprise.

Learn how Colab Enterprise authentication uses lockdown credentials to access GCS buckets and resources with your login, or switches to GCP-managed runtimes when disabled, requiring service accounts or ADC login.

Explore workbench in Vertex AI, create and customize an instance, choose machine types and JupyterLab versions, and learn access control and service account best practices.

Explore the GCP Vertex AI workbench and JupyterLab, using a Compute Engine VM to access BigQuery, clone Vertex AI samples from GitHub, and train models with TensorFlow or PyTorch.

Create schedules and monitor executions in the workbench, selecting machine type and environment, and configure five-minute intervals with cron or scheduling options to store outputs in a bucket.

Create and monitor a schedule that runs a notebook as a custom training job every five minutes. First execution initializes the environment, then subsequent runs proceed normally.

Explore how to build, train, and deploy ML models on GCP using Vertex AI and services. Learn data storage options, training approaches, deployment strategies, and Pub Sub Dataflow for streaming.

Compare cloud storage and database options for machine learning workflows, from object storage to BigQuery and Firestore, then evaluate training and deployment options across Vertex AI services.

Explore Vertex AI services by linking data to a GCS bucket or BigQuery, train models with AutoML or custom options, register them in the model registry, and deploy via endpoints.

Learn to create and configure datasets in Vertex AI, choosing region and data type (image, tabular, text, video) for supervised or unsupervised training; text and video datasets are deprecated.

Train a regression model to predict house prices from features like size, bedrooms, bathrooms, age, and neighborhood quality, using a CSV dataset uploaded to a GCS bucket.

Create a Vertex AI dataset for tabular regression, upload the CSV to a GCS bucket, import the data, generate statistics, and prepare for model training.

Train a new AutoML regression model using your dataset to predict house price from features like bedrooms, bathrooms, and size, with automatic data type detection and a random 80/10/10 split.

Deploy your trained Vertex AI model to an endpoint, export the saved model to a bucket, split traffic between versions, and enable monitoring and explainability to detect inference drift.

Test your deployed model via rest or python, using a json input and curl in the cloud shell to obtain a house price prediction with a lower and upper bound.

Import a dataset from a BigQuery table, create a US central region dataset, auto detect the schema, and train a regression model with AutoML by selecting the target column.