Kubernetes at multi-tenant scale, and the isolation and autoscaling layers on top.
I'm a PM who reads controller code. Not to write it, but because the product decisions in this domain are inseparable from the mechanics — you can't have a useful opinion about an autoscaling policy without understanding what the scheduler will actually do with it.
| Technology | Depth | What I did with it |
|---|---|---|
| Kubernetes | Expert | CKA certified. Autoscaling, namespace operations, multi-tenant platform design. |
| Karpenter | Expert | Provisioning and consolidation behaviour under seasonal load. |
| VPA / HPA | Expert | Right-sizing policy and CRD-based controller design. |
| gVisor | Practiced | Syscall-level isolation for untrusted agent workloads. |
| Cilium eBPF | Practiced | Network policy enforcement at the sandbox boundary. |
| AWS EKS | Practiced | Primary production environment. |
| Istio | Familiar | Service mesh evaluation and traffic policy context. |
| GCP GKE | Familiar | Comparative evaluation against managed alternatives. |
Kubernetes won the orchestration argument and lost the usability one. The durable product opportunity in this domain is not another abstraction over it — it's making the existing primitives legible to the people who have to operate them at 3am.