Sunil Mehta / infra
Open to remote roles · Gurugram
pick one to reshape the dots

I’m Sunil, Founding Lead DevOps & MLOps Engineer at ORGN. For 9+ years I have built and run the platforms AI products depend on, from multi-cloud Kubernetes and confidential GPU compute to secure delivery and cost governance, so that what runs on them never has to think about them.

Explore my work ↗ Resume ↓

Merged upstream in Argo CDMLflowKAI Schedulerllm-d

Scroll
Running in production300+ TDX GPU nodes
Building acrossCloud, security & AI
Thinking aboutAutoscaling & supply chains
  • Kubernetes
  • Intel TDX
  • Argo CD
  • Terraform
  • vLLM
  • KServe
  • Kueue
  • Volcano
  • KAI Scheduler
  • MLflow
  • Kubeflow Pipelines
  • Prometheus
  • Grafana
  • OpenTelemetry
  • Kyverno
  • cosign
  • SLSA provenance
  • Istio
  • Helm
  • GitHub Actions
  • Go
  • Rust
  • Python
  • GCP
  • AWS
  • Azure
300+Intel TDX confidential GPU nodes in production
100M+Requests per month through Rust LLM gateways
99.95%Uptime across 200+ services and 10+ clusters
55%Hosting cost cut with zero-downtime migrations

How a change reaches production

The platform map →
  1. CommitGitHub Actions signs in through Workload Identity, with no stored keys
  2. ScanSecrets, code, dependencies, images and infrastructure code
  3. SignImages signed with cosign in Artifact Registry
  4. SyncArgo CD reconciles every cluster from Git
  5. AdmitKyverno and OPA refuse what breaks policy
  6. RunGKE clusters and 300+ TDX GPU nodes
  7. WatchFalco at runtime; Prometheus, Loki and Grafana

The path on the platform I run at ORGN.

One command, one root cause

The tools →

Two of the tools are command lines. One reads a broken namespace and names the cause; the other reads Prometheus and prices the GPUs nobody used.

  • Read-only: neither changes anything in the cluster.
  • Every finding comes with its evidence and the fix.
  • Both outputs here are real runs, shortened.
$ kubectl diagnose -n kd-liveness-wrong-port1 problem: 1 warning. 1. WARNING  Liveness probe kills the container  KD112   Deployment kd-liveness-wrong-port/web   Cause: container web fails its liveness probe (GET / on port 9090, but the          container declares port 8080), so the kubelet restarts it (5 restarts)   Evidence:     [1] Pod …/web-67fbcb44db-snwhf: event Unhealthy (6 times)         │ Liveness probe failed: … dial tcp 10.244.0.26:9090: connect: connection refused         $ kubectl get events -n kd-liveness-wrong-port --field-selector …     [2] Pod …/web-67fbcb44db-snwhf: event Killing (6 times)         │ Container web failed liveness probe, will be restarted   Fix: Point the probe at a port and path the app serves, or give it a startupProbe        or longer initialDelaySeconds if it starts slowly.

kube-diagnose on the fault suite’s recording of a wrong liveness port; gpucost on its synthetic demo fleet. Some lines are left out, and “…” marks a cut within a line.

30-second summary
Keyboard shortcuts

Shortcuts are off while you type in a field.

↑ ↓ move↵ openesc close