Situation: At Airinmar I managed Kubernetes workloads split across AWS EKS for cloud-native services and on-premises clusters for legacy applications still tied to internal data sources. The teams supporting these were under pressure because every deployment to either environment required manual coordination.
Task: I needed to standardise deployment across both environments and accelerate the release cadence without sacrificing reliability.
Action: I rolled out ArgoCD as the GitOps controller for both clusters with Helm charts as the deployment unit. I containerised the legacy internal applications and brought them under the same GitOps workflow. I built GitHub Actions CI pipelines that built, tested, and pushed images to ECR, with ArgoCD reconciling the changes. I also managed the underlying VMware ESXi, Horizon VDI clusters, firewalls, and switches that supported the on-prem environment.
Result: Delivery time was halved. Deployments became repeatable and audit-trailable, and both engineering teams could release independently without coordinating manually.
Preventative action: The GitOps model meant rollback was instant — any bad release could be reverted with a Git revert, no manual cluster intervention. That removed a class of incident we'd been seeing previously.