whoami
kubectl get engineer namit -o wide
NAME ROLE ORG EXP STATUS
namit DVP · Sr. Architect, SRE kotak-mahindra-bank 8y Running
I build and run the digital platform behind Kotak's Mobile Banking 2.0 — banking infrastructure that millions of customers rely on and never have to think about.
99.99% availability · 8.0 years in production
I'm Namit — a platform engineer who has spent eight years turning fragile release nights into boring ones. I lead the Digital Platform Team for Mobile Banking 2.0 at Kotak Mahindra Bank: the clusters, the pipelines, the mesh, and the people who keep them healthy.
Automate yourself out of toil.
If a human did it twice, it's a script. If a script ran twice, it's a pipeline. The best on-call shift is the one where nothing pages you.
You can't AI what you can't observe.
Every AIOps ambition dies without clean telemetry underneath it. Instrument first, correlate second, automate third.
Every skill here is wired to the others the way it is in production. Hover to trace the connections; click to pin one.
Written the way incidents are: problem, action, impact. Metrics marked TODO: are being de-classified from war-room notes.
East–west traffic between banking microservices was unencrypted and invisible, and a single shared ingress mixed internal and external traffic paths — unacceptable for a regulated workload.
Designed and shipped a production Istio service mesh on EKS with mTLS anchored to AWS Private CA, fronted by a Kong dual-gateway ingress that hard-separates internal and external traffic.
100% of service-to-service traffic now rides mTLS with automated cert rotation. TODO: services onboarded, p99 latency overhead, cert-rotation incidents (zero so far).
Logs lived and died on individual nodes. Debugging a cross-service failure meant kubectl-exec archaeology across dozens of pods under incident pressure.
Rolled out a Fluent Bit DaemonSet pipeline into OpenSearch — structured parsing, Kubernetes metadata enrichment, index lifecycle management sized for bank-scale volume.
One query now answers what used to take an hour of pod spelunking. TODO: GB/day ingested, MTTR delta, retention window.
Alert volume outgrew human triage, and release risk was judged by gut feel — in an environment where a bad call is a regulatory event, not just a bad day.
Authored the bank's 12-month AI-infrastructure roadmap: alert-noise reduction via correlation, ML-scored release risk gates in the pipeline, and predictive SLO-breach detection on golden signals.
Roadmap approved and funded; first phase in flight. TODO: % alert reduction in pilot, releases scored, prediction lead-time.
Every team had its own half-configured dashboards and nobody trusted any of them. Incidents began with an argument about whose graph was right.
Stood up a unified stack — Prometheus metrics with SLO burn-rate alerts, golden-signal Grafana dashboards per service, Jaeger tracing through the mesh, Kiali for topology.
One agreed-upon picture of production. War rooms start with data, not debate. TODO: services covered, alert precision before/after.
Regulatory DR requirements demanded provable recovery — not a document that claims failover works, but a failover that actually has.
Architected the DR/BCP posture for the platform's banking workloads: multi-region topology, data replication strategy, runbooked failover, and scheduled game-day drills.
Recovery is rehearsed, timed, and evidenced for audit. TODO: measured RTO/RPO, drill cadence, last failover duration.
SEV-1 = deep dive · SEV-3 = quick tip. Runbooks are written after the pager stops.
Evaluating me for a Principal SRE, Platform Architect, or engineering-management role? Open an incident — I acknowledge fast.