Helm Starter Kits & Internal Developer Platform
Created opinionated service templates with built-in HPA, readiness probes, OpenTelemetry tracing, and sealed secrets.
SaaS Paved-Road EKS Platform
Empowering 14 distributed engineering squads with a self-service Kubernetes platform, ArgoCD GitOps pipelines, Karpenter instant autoscaling, and Kyverno policy guardrails.
1 Day
Dev squad onboarding (was 3 wks)
45 sec
Node autoscaling time
42%
EC2 compute bill reduction
100%
Declarative GitOps coverage
Platform Architecture Blueprint
GitHub PR Developer Code Commit
→
ArgoCD Engine Automated Reconciliation
→
EKS Control Plane Multi-Tenant Namespaces
→
Karpenter Provisioner Spot + Graviton Fleet
The Challenge
The company grew rapidly from 2 to 14 feature teams. Developers were waiting weeks for infrastructure provisioning, while heterogeneous snowflake clusters created security drift, high cloud bills, and weekend outage fire drills.
Slow Delivery Velocity
Engineers had to write raw Kubernetes manifests from scratch, often failing security audits due to missing resource limits and security contexts.
Inefficient Compute Waste
Old Cluster Autoscaler took 4–7 minutes to react to spike traffic, forcing teams to run 300% overprovisioned on-demand instances overnight.
The Architectural Solution
Karpenter Dynamic Just-in-Time Nodes
Replaced legacy Node Groups with Karpenter. Workloads automatically bin-pack onto AWS Graviton (ARM64) and Spot instances, provisioning compute in 45 seconds.
Kyverno Admission Controllers & Cilium Network Policies
Enforced zero-trust namespace isolation, non-root containers, and automated image vulnerability blocking before pods reach cluster scheduling.
Technologies Employed
AWS EKS 1.30 Karpenter AWS Graviton3 ArgoCD Kyverno Cilium eBPF Terraform Prometheus / Thanos