Suraj Kumar

The security team got vulnerability reports as PDFs every quarter. By the time anyone read them, the data was already stale. Nobody could answer a simple question: is CVE-2024-XXXXX running in production right now?

I built a Python service that pulls container scan results and joins them against what is actually scheduled in the cluster — not what is in the registry, what is running. A critical CVE in an internal batch job gets deprioritised. A medium CVE in an internet-facing pod gets flagged immediately.

We were a 4-person platform team supporting multiple client engagements. Every day: "Can you create a namespace?" "Can you scale this up?" "I need to see the logs for pod-whatever."

I built a FastAPI service that let product teams do this themselves. One API across EKS, AKS and GKE. Quotas and guardrails baked in — if you try to request 100 CPU cores, it says no before anything gets created.

Nomad and Puppet failures were invisible until something downstream broke. By then you were already in incident mode.

I wrote a Python monitor that polls the scheduler and config management layer directly — the services everything else depends on. Not the apps. The plumbing. Output goes straight into the alerting pipeline we already respond to. No new dashboard nobody watches.