Google Cloud Platform Training in Pune with Placement

Cloud computing models, Google Cloud's global infrastructure (Regions, Zones, Edge points of presence and the private backbone), and the resource hierarchy — Organisation, Folders, Projects, Resources — which is the structure that IAM, billing and policy all inherit from.

Every student finishes week 1 with a project, billing configured against free-trial credit, budget alerts wired up, and both the `gcloud` CLI and Cloud Shell working. Cost gets a dedicated session up front, because an idle GKE cluster will consume trial credit faster than anything else in this course.

Google Cloud IAM is structurally different from AWS and Azure, and the difference is where most cross-cloud engineers make mistakes. Principals, roles (basic, predefined, custom) and the allow-policy model; how policies inherit down the resource hierarchy and why a grant at folder level is rarely what someone intended.

Service accounts get extended treatment — impersonation, key-free authentication, and Workload Identity Federation for CI/CD and for GKE pods. Organisation policy constraints and the principle of least privilege close the module.

Google Cloud's VPC is global rather than regional, which changes network design in ways worth understanding explicitly. Auto-mode versus custom-mode VPCs, subnets and secondary ranges for GKE, firewall rules with network tags and service accounts as targets, and Cloud NAT with Cloud Router for private egress.

Cloud Load Balancing is taught by tier and protocol — global external HTTP(S) with Cloud CDN, regional internal, and the proxy versus passthrough distinction. Shared VPC, VPC Peering, Private Service Connect and Cloud DNS complete the module.

The VM layer. Machine families and custom machine types (a genuine Google Cloud advantage worth knowing how to price), persistent disk types, images and snapshots, startup scripts and metadata, plus OS Login for SSH access governed by IAM rather than by keys scattered across a team.

Spot and preemptible instances follow, then Managed Instance Groups — instance templates, autohealing with health checks, regional distribution, rolling updates and autoscaling — ending with a service that survives deliberate instance deletion.

Google Kubernetes Engine, taught with Autopilot as the 2026 default and Standard mode as the escape hatch when you need node-level control. Cluster creation, VPC-native networking with the secondary ranges planned in week 3, node pools, and cluster autoscaling.

The GKE-specific material is the point of the module: Workload Identity for pod-level IAM without keys, Ingress and Gateway API with Google Cloud load balancers, Config Connector, and the operational differences that make an Autopilot cluster cheaper to run but harder to customise.

Serverless on Google Cloud, which is the platform's strongest differentiator for small teams. Cloud Run for containers — concurrency (which is genuinely different from Lambda's model), scale-to-zero, minimum instances for latency, revisions and traffic splitting for canary releases, and VPC connectors for private backend access.

Cloud Run Jobs handle batch work, Cloud Functions covers event-driven glue, and Eventarc plus Pub/Sub provide the event backbone. The lab compares the same workload deployed to Cloud Run and to GKE on cost and latency.

Cloud Storage in depth — buckets, the four storage classes and the retrieval and early-deletion charges that make Archive cheap only under the right access pattern, lifecycle rules, object versioning, retention policies and Bucket Lock.

Access control gets a full session because it is the most common source of accidental public exposure: uniform bucket-level access versus fine-grained ACLs, signed URLs, and public-access prevention. Customer-managed encryption keys, Filestore and Persistent Disk round out the module.

The managed database estate. Cloud SQL for PostgreSQL, MySQL and SQL Server — high availability, read replicas, automated backups and point-in-time recovery, private IP and the Cloud SQL Auth Proxy.

Cloud Spanner is taught for what makes it distinctive — horizontal scale with strong consistency, and the interleaving and primary-key design that avoid hotspots. Firestore covers document modelling and its query constraints, and Memorystore provides managed Redis for the caching patterns used earlier in the course.

BigQuery is the service that most often decides whether a Pune GCP candidate stands out, so it gets a full module rather than a passing mention. The separation of storage and compute, on-demand versus capacity pricing, and why an unoptimised query is a billing event rather than merely a slow one.

Partitioning and clustering, table design, nested and repeated fields, materialised views, scheduled queries, and reading the query execution plan to find the expensive stage. External and BigLake tables, and BigQuery ML for in-warehouse models, close the module.

Provisioning everything built so far from code. Terraform 1.7+ with the `google` and `google-beta` providers, remote state in a GCS bucket with locking, modules, and workspaces for environment separation.

Google-specific practice follows: the Cloud Foundation Toolkit modules, project-factory patterns for organisations that create projects continuously, and importing resources created through the console. `terraform plan` is treated as a review artefact, and drift detection is run against a deliberately hand-modified resource.

Delivery pipelines on Google Cloud. Cloud Build triggers, build configuration, substitutions, private pools and build-time secrets from Secret Manager; Artifact Registry for container images and language packages, with vulnerability scanning enabled.

GitHub Actions using Workload Identity Federation covers the keyless external pattern. Cloud Deploy handles progressive delivery through dev, staging and production targets, and the module ends with rollout, approval and rollback wired into the Cloud Run and GKE services built earlier.

Google Cloud's operations suite. Cloud Logging with log buckets, sinks and exclusion filters (the lever that controls logging cost), the Logs Explorer query language, and log-based metrics that turn a log pattern into an alertable signal.

Cloud Monitoring covers metrics, uptime checks, dashboards, alerting policies and notification channels, plus SLO monitoring with error budgets. Cloud Trace, Cloud Profiler, OpenTelemetry instrumentation and Managed Service for Prometheus complete the module.

Security consolidated across the whole estate. Cloud KMS with key rings, rotation and CMEK applied to storage and databases; Secret Manager versioning and access; Cloud Armor for WAF and DDoS; VPC Service Controls for data-exfiltration perimeters; and Security Command Centre findings.

The cost half covers billing export to BigQuery, label-based cost allocation, committed use discounts, and the standard waste list. Each student then reviews their capstone against the Google Cloud Architecture Framework pillars.

Google Cloud's AI platform, which is a frequent differentiator in Pune GCP interviews. Vertex AI Workbench for managed notebooks, Model Registry and endpoints for serving, and Vertex AI Pipelines for repeatable training workflows.

The generative half covers Model Garden including Gemini and open-weight models, grounding and Vertex AI Search for managed retrieval-augmented generation, Agent Builder for tool-using assistants, and embeddings with vector search. The lab ships a small grounded RAG service end to end.

Full-time capstone work followed by structured interview preparation. You build a complete Google Cloud system — Terraform-provisioned infrastructure, a Cloud Run or GKE workload, a Cloud Build pipeline, BigQuery analytics, and monitoring with an alerting policy — documented well enough that a reviewer can deploy it themselves.

Interview preparation is run as three rounds matching Pune GCP hiring: a BigQuery query-optimisation round, an architecture and scenario round, and a troubleshooting round. Resume, LinkedIn and GitHub polish included.