Platform Engineer · Chicago, IL
I'm a platform engineer in Chicago. Over the past five years I've run infrastructure for two very different kinds of consumer traffic: a high-volume app with steady load around the clock, and a ticketing marketplace where an on-sale can multiply load in minutes. The goal was the same in both: make it uneventful. That means infrastructure declared in code, deployments where the merge is the only human step, capacity that follows load on its own, and observability that takes on-call from a spike on a graph to the exact request.
A growing part of that work is building AI agents for the teams around me. Platform and product engineers now use them every day to catch policy violations and debug incidents.
AWS SAA
CKAD9 clusters · 16 teams
A GitOps delivery path across multiple AWS accounts: teams ship by editing Helm values and merging — Argo CD does the rest.
How
Problem — deploys ran through tickets and hand-offs, and every team paid for the wait.
Built — desired state in git, Argo CD reconciling it continuously, and Helm values as the only interface a team has to touch.
Result — 16 teams deploy on their own across 9 clusters; the merge is the only human step.
Capacity in both directions
Dynamic node provisioning with Karpenter and queue-driven worker scaling with KEDA — capacity that follows load up and back down.
How
Problem — on-sale traffic arrives in bursts far above baseline, and capacity has to be there before the queue backs up.
Built — KEDA scales workers on queue depth; Karpenter provisions nodes the moment those workers have nowhere to schedule, and consolidates them when the rush passes.
Result — spikes absorbed without a human in the loop, and spend that tracks real use.
Audit → enforce
Kyverno guards the cluster; an AI agent does the work between audit mode and enforce mode — tracing violations to their source in git and building the case for each flip.
How
Problem — admission policies start in audit mode and tend to stay there: violations pile up unread, and the enforce flip is deferred because proving it safe costs an engineer-day nobody burns.
Built — an agent that traces each violation to the chart or values file that produces it, triages it into one of four buckets — fix the workload, exempt with a paper trail, fix the policy, or blocked upstream — ships the workload-fix PRs, and assembles the evidence case for each enforce flip.
Result — enforcement earned incrementally, with evidence. The agent never flips a policy itself; a human reviews the case and merges.
git revert.
Plants at home, time outside whenever I can get it — ideally with a sunset.
I like wandering antique stores — someone's old is my new.
Just for myself, not for an audience. At the moment it's a ukulele.
Open to platform, infrastructure, and DevOps work. If you'd like to talk, my inbox is open.