Infrastructure Engineer · growing into SRE / DevOps, with an AIOps edge · Seoul, Korea

A social-work → infrastructure career switch. Everything below I've built or operated hands-on — and where something is only code, or only installed, I say so.

An infra engineer who actually runs agentic AI in his daily workflow — not just reads about it. (Claude Certified Architect – Foundations)

• 🔧 Hands-on infra & systems ops — HPE/Dell servers, Cisco networking, M365 — for global enterprise clients via a global IT delivery chain. Now automating the repeatable parts.
• ☁️ Cloud / Platform — AWS + EKS provisioned and operated as Terraform code, with remote state and OIDC-authenticated CI (see ⭐ below)
• 📊 Observability — a custom exporter feeding Grafana Cloud dashboards, alerts, and an SLO with an error budget — all managed as code, on a site I actually run
• 🧪 Production automation with DevOps principles — idempotent reconciliation, backoff+jitter retry honoring 429/503, audit logging
• 🤖 Infra × AI — multi-agent (Claude Code) pipelines I run day to day
• 🎓 Social work → IT — I bring an ops-and-people lens to incidents and handoffs, and I learn in public

An AWS EKS platform defined as Terraform code — VPC / EKS / managed node group / IRSA / RDS — provisioned on my own account, operated, and torn down deliberately for cost control.

• Remote state on S3 (versioning · SSE · public-access block) with DynamoDB locking
• GitHub OIDC — CI runs terraform plan on every PR with no long-lived AWS keys in the repo. The trust policy pins sub to this repo's pull_request and main (no wildcards), and the CI role has no write permissions at all — not even s3:PutObject — so apply and state corruption are structurally impossible.
• Cost forensics from the real bill — 63% of one month's charges turned out to be EKS extended-support surcharge, not resource usage. That's money spent on not upgrading, so I pinned the cluster to a standard-support version.
• A verification table in the README that separates what was apply-verified from what is code-only. Building it flipped two of my own conclusions: a resource missing from state that billing proved had run 38 hours, and a component helm reported as deployed that had never actually created anything. One class of evidence would have gotten both backwards.

Not in this repo: Karpenter, CloudFront, Container Insights — the README says so explicitly, and ArgoCD / the load balancer controller are marked installed only.

• 📖 Working toward CKAD
• 🔜 Next up — a log axis (Loki) for the observability stack, and checkov static analysis in the Terraform CI

Claude Certified Architect – Foundations · AWS Solutions Architect Associate · AWS Developer Associate · GCP Associate Cloud Engineer · Oracle OCI Architect Associate · Cisco CCNA · HPE Hybrid Cloud · Azure AZ-900 · Google IT Support — CKAD in progress