Hire senior Platform engineers in your time zone.

A senior platform engineer owns the cluster as shared infrastructure, not just the manifests for one service. That includes namespace design, RBAC policies that keep teams from stepping on each other, and resource quotas that prevent one workload from starving the rest of the cluster.

They own the deployment path: Helm charts or Kustomize overlays that make shipping a new version of a service repeatable and safe, with sensible defaults for health checks, resource requests, and rollout strategy so a bad deploy does not take down the whole cluster.

Autoscaling is their responsibility too, tuning horizontal pod autoscaling and cluster autoscaling so workloads have the capacity they need without paying for idle nodes around the clock.

When something breaks, they are the ones diagnosing it: a pod stuck in a crash loop, a networking policy blocking traffic it should allow, or a node running out of resources. Observability, Prometheus for metrics, Grafana for dashboards, structured logs for the rest, is what makes that diagnosis fast instead of a guessing game.

Helm is close to a default for packaging and deploying applications on Kubernetes, and a candidate should be comfortable writing and maintaining charts, not just installing ones someone else wrote. Kustomize is a common alternative or complement, especially for managing environment-specific overlays.

Observability tooling matters as much as the cluster itself. Prometheus and Grafana are the standard pairing for metrics and dashboards, and a candidate should know how to instrument an application, not just read an existing dashboard.

Most production clusters run on a managed offering, EKS, GKE, or AKS, rather than self-hosted from scratch, and familiarity with your specific provider's networking and IAM model saves real ramp-up time. Terraform frequently provisions the cluster and surrounding infrastructure alongside it.

Kubernetes has also become the common way to schedule GPU-backed inference workloads for AI applications, alongside the rest of a company's services, so candidates working in organizations running machine learning in production increasingly need to understand GPU scheduling and resource allocation on top of the fundamentals.

Ask about a real production incident on Kubernetes: a pod that would not start, a service that could not reach another service, or a node running out of resources. The way someone walks through diagnosis, checking events, logs, and resource metrics in order, tells you far more than a list of kubectl commands.

Ask them to walk through a Helm chart or set of manifests they maintain, and why specific resource requests, limits, and health check settings were chosen. Good answers reference the actual workload's behavior, not generic defaults copied from a tutorial.

Probe RBAC and namespace design directly if your cluster is or will be multi-team. Ask how they would set up a new team on a shared cluster without giving them more access than they need.

BetterEngineer already runs this kind of evaluation, incident walkthroughs, manifest and Helm chart review, and RBAC design questions, before a candidate ever reaches your calendar, so the profiles you see have already cleared this bar.

Quick evaluation checklist:

• Ask about a production incident they handled on Kubernetes and how they diagnosed it
• Review a Helm chart or set of manifests they have maintained for a real service
• Check their approach to resource requests, limits, and autoscaling configuration
• Probe RBAC and namespace design for a multi-team or multi-tenant cluster
• Assess hands-on experience with a managed offering, EKS, GKE, or AKS, not just local Minikube