Dinesh Sai ← All projects

Case study · Platform engineeringProduction, MedTech

A Kubernetes platform where every rollout is traceable.

Production Amazon EKS for HemoCue's medical diagnostics software and SCIEX's scientific instrument software at Danaher. GitOps with Argo CD, Helm charts for every service, autoscaling, and SRE practice that keeps 50+ production services at 99.9% uptime.

  • Role: Lead DevOps Engineer
  • Amazon EKS · Argo CD · Helm
  • SRE · incident response
apps/notes-dev.yaml (public lab)
apiVersion: argoproj.io/v1alpha1
kind: Application
spec:
  source:
    repoURL: …/devops-gitops-lab.git
    path: infra
  syncPolicy:
    automated:
      prune: true     # delete what git deleted
      selfHeal: true  # undo manual drift
From a public lab repository: git is the only way to change the cluster.
5M+requests a day
99.9%uptime, 50+ services
20+microservices on Helm
40%lower MTTR

Results from my work at Danaher, as listed on my resume.

The problem

Diagnostic workloads can't be "mostly up".

The services behind medical diagnostics are critical, regulated, and released on a strict governance cycle. The platform has to absorb load spikes without manual scaling, roll changes out and back safely, and answer the auditor's question of exactly what ran in production and when.

The approach was GitOps plus SRE discipline: the cluster's desired state lives in git, every service ships the same way through Helm, and reliability is measured against SLOs instead of judged by feel.

What I built and run

The platform, layer by layer.

  1. GitOps delivery

    Argo CD with Kustomize gives auditable, environment-specific rollouts. Every change to production is a git commit, with full traceability and one-click rollback.

  2. One chart pattern for every service

    Reusable Helm charts for 20+ microservices, with environment-specific overrides, rollback strategies and automated readiness and liveness checks. A new service gets production-grade deployment behaviour by default.

  3. Autoscaling

    HPA and VPA keep capacity matched to demand. EKS autoscaling serves 5M+ requests a day without manual intervention.

  4. Safe releases

    Rolling deployments are the default, with blue-green patterns for the releases where risk has to be close to zero.

  5. Observability and response

    The ELK Stack centralises logs across microservices, and Grafana and CloudWatch cover metrics. Together with incident runbooks, this cut mean time to recovery by 40%. SLOs and SLAs are tracked for the critical diagnostic workloads.

  6. Recovery

    Multi-AZ backup and restore for stateful workloads on EKS and RDS, tested with disaster-recovery simulations.

Before Danaher

How the experience built up.

Subscribe IT · fintech

Operated production Kubernetes clusters, improving uptime SLAs by 20%. Rolled out Grafana, Prometheus and Node Exporter with Ansible across 100+ servers, cutting manual setup time by 70%.

CSS Corp and MNP Technologies

Monitored 10+ Kubernetes systems through Grafana, ran Docker Swarm clusters, managed production Kubernetes deployments, and provided 24/7 support across 5+ teams.

In code

Related work you can look at.

Employer clusters are confidential. These repositories show the same patterns.

Stack

Tools used

Want the details?

I can talk through GitOps structure, chart design and SLO practice in more depth, within what my employer allows.

Get in touch More projects