Case Studies

Kubernetes Cost & Reliability Overhaul

A 40-node EKS cluster ran on static, hand-picked node groups with no autoscaling discipline, and pods were getting OOMKilled during every traffic spike. Compute spend kept climbing while reliability got worse, not better. Rightsized every workload's requests from real Prometheus usage data, then replaced static node groups and Cluster Autoscaler with Karpenter for just-in-time, bin-packed provisioning. Added PodDisruptionBudgets so consolidation never dropped a service below its minimum replica count.

AWS · EKS · Karpenter · Terraform · Prometheus

SOC 2-Ready Kubernetes Security Hardening

Every pod inherited a shared node-wide IAM role with access to S3 buckets it had no business touching, the cluster network was flat with zero segmentation, and CI pipelines held long-lived static credentials in plaintext environment variables. The client's SOC 2 pre-audit failed on all three. Scoped IAM to the ServiceAccount level with IRSA, rolled out default-deny NetworkPolicies with OPA Gatekeeper enforcing them at admission, and migrated every CI credential to Vault-issued, short-lived leases with full audit logging.

AWS · EKS · IRSA · OPA Gatekeeper · Vault

GitOps Pipeline Rebuild Before Peak Season

Deploys took 45 minutes and required a human to SSH in and run a script, so the team shipped once a week at most. With Black Friday six weeks out, a bad release meant an hours-long manual rollback with no tested procedure. Rebuilt CI around dependency and Docker layer caching, then replaced the manual release process with ArgoCD-managed GitOps: every environment syncs from Git automatically, drift self-heals, and rollback is a single revert.

GitHub Actions · ArgoCD · Helm · Kubernetes · Docker

Zero-Downtime Multi-Region Database Failover

The DR plan existed only on paper: backups ran nightly but had never once been restored, and the whole platform sat in a single AWS region. A compliance review flagged both as unacceptable ahead of a hospital-network contract renewal. Stood up an asynchronous streaming replica in a second region with Patroni-managed promotion, wired automated Route53 health-check failover, and added a nightly job that restores the latest backup to a throwaway instance and verifies row counts before anyone's morning coffee.

AWS · PostgreSQL · Terraform · Route53 · Patroni