Hands-on Lab — GAIL Google Cloud Generative AI Leader

Overview

Prerequisites

Steps

1. 1.Provider, project services, naming

   Enable Vertex AI, Discovery Engine (Vertex AI Search), Cloud Notebooks, and Cloud Monitoring APIs.

   Copy

   terraform {
     required_version = ">= 1.5"
   
     required_providers {
       google = { source = "hashicorp/google", version = "~> 6.0" }
     }
   }
   
   provider "google" {
     project = "your-project-id" # REPLACE
     region  = "us-central1"
   }
   
   locals {
     labels = {
       project    = "certlabpro-gail"
       managed_by = "terraform"
     }
   }
   
   resource "google_project_service" "aiplatform" {
     service            = "aiplatform.googleapis.com"
     disable_on_destroy = false
   }
   
   resource "google_project_service" "notebooks" {
     service            = "notebooks.googleapis.com"
     disable_on_destroy = false
   }
   
   resource "google_project_service" "discoveryengine" {
     service            = "discoveryengine.googleapis.com"
     disable_on_destroy = false
   }
   
   resource "google_project_service" "monitoring" {
     service            = "monitoring.googleapis.com"
     disable_on_destroy = false
   }

2. 2.Provision a Vertex AI Workbench instance for prototyping

   Provisions:

   • Vertex AI Workbench

   Vertex AI Workbench is the Jupyter-in-the-cloud surface every GenAI prototyping flow on GCP starts from — equivalent to Amazon SageMaker Studio / Azure ML Workbench. GAIL exam tests this as the go-to data-scientist seat primitive.

   We provision a small e2-standard-2 Workbench instance pre-loaded with the GenAI Python libraries. The Workbench runs in a managed Compute Engine VM under the hood — you'll see it in the GCE console too. Stop the instance via the Vertex AI Workbench console when not actively using it to avoid ~$50/month while idle.

   Copy

   resource "google_workbench_instance" "main" {
     name     = "certlabpro-gail-workbench"
     location = "us-central1-a"
   
     gce_setup {
       machine_type = "e2-standard-2"
   
       boot_disk {
         disk_size_gb = 100
         disk_type    = "PD_STANDARD"
       }
     }
   
     labels = local.labels
   
     depends_on = [google_project_service.notebooks]
   }

3. 3.Create a Discovery Engine data store (Vertex AI Search grounding source)

   Provisions:

   • Vertex AI Search

   Discovery Engine (formerly Vertex AI Search) is the GAIL exam's flagship RAG primitive — point it at a corpus (GCS bucket, website crawl, BigQuery dataset, or structured JSON), it indexes the content, and downstream Gemini calls can ground their answers in that corpus.

   We create a data store of type GENERIC — the conceptual container for indexed content. Production deployments populate the store with documents via the Discovery Engine API; for the lab, the empty store is the demonstration that the resource exists and bills accordingly.

   The GAIL exam tests this data store → search app → grounded answers triangle as the standard enterprise-RAG-on-GCP shape.

   Copy

   resource "google_discovery_engine_data_store" "main" {
     data_store_id     = "certlabpro-gail-store"
     display_name      = "GAIL lab data store"
     location          = "global"
     industry_vertical = "GENERIC"
   
     content_config = "CONTENT_REQUIRED"
     solution_types = ["SOLUTION_TYPE_SEARCH"]
   
     depends_on = [google_project_service.discoveryengine]
   }

4. 4.Alert on Vertex AI prediction error rate

   Provisions:

   • Cloud Monitoring

   GenAI workloads have an operational dimension the GAIL exam keeps surfacing: prediction errors, latency, cost-per-token. We wire a Cloud Monitoring alert on Vertex AI's prediction error metric — fires when the error rate exceeds 5% over 5 minutes.

   With four blocks in place (provider+APIs, Workbench for prototyping, Discovery Engine data store for grounding, and a Cloud Monitoring tripwire for operational health), the GAIL conceptual landscape is mapped to actual provisioned infrastructure. Real GenAI deployments layer model garden picks, custom training, prompt management, agent builder, and Model Armor / Responsible AI controls on this foundation — but the substrate above is the shape you'll see in GAIL exam scenarios.

   Copy

   resource "google_monitoring_alert_policy" "vertex_prediction_errors" {
     display_name = "GAIL lab — Vertex AI prediction error rate"
     combiner     = "OR"
   
     conditions {
       display_name = "Prediction error rate > 5% over 5 minutes"
   
       condition_threshold {
         filter          = "metric.type=\"aiplatform.googleapis.com/prediction/online/error_count\" AND resource.type=\"aiplatform.googleapis.com/Endpoint\""
         duration        = "300s"
         comparison      = "COMPARISON_GT"
         threshold_value = 0.05
   
         aggregations {
           alignment_period   = "60s"
           per_series_aligner = "ALIGN_RATE"
         }
       }
     }
   
     # notification_channels = []  # add channels via console or separate TF resource
   
     depends_on = [google_project_service.monitoring]
   }

Cleanup