DevOps engineer. I take complex systems into production: Kubernetes, Terraform, and CI/CD across Azure and GCP, embedded directly in client platform teams until it ships. The good kind of gravity: someone else's problem, lifted off their shoulders.
I build the unglamorous plumbing beneath other people's magic. Clusters that scale without panic. Pipelines that deploy without ceremony. The good kind of infrastructure, the kind you stop noticing.
A loose timeline: from software engineer to forward-deployed DevOps, embedded in client platform teams across finance, pharma, and automotive.
Embedded in BMW's platform team, leading a major modernization of how monitoring runs on managed Kubernetes, and owning the reliability of what reaches production.
Ran observability for a portfolio of regulated financial clients, bringing fragmented monitoring under one dependable platform and putting AI into the on-call workflow where it made triage smarter.
Owned the delivery pipeline behind a healthcare ordering platform, and acted as the engineering bridge to the client's technology leadership.
Kept Kubernetes workloads and data migrations shipping reliably for an automotive platform, with less downtime along the way.
Grew from software engineer into DevOps, taking ownership of CI/CD and Kubernetes for cloud-native delivery across client projects.
Eight concentric systems. Outer rings I design in my sleep; inner ones I still pair with the docs.
Production-ready GitOps with automated image updates. ArgoCD Image Updater, Kubernetes HPA with multi-metric autoscaling (CPU + Pub/Sub), External Secrets Operator, Slack notifications for deployment tracking.
Dev, staging, prod on GCP. DRY with Terragrunt: GKE, VPC, Cloud CDN, CI/CD via GitHub Actions.
Full-stack microservices: Auth, Workout, Analytics. Python/Flask on GKE via Terraform + Terragrunt, Postgres, NGINX, ELK for observability.
Also: orphaned Azure disks quietly burning $187/mo. Why GKE production wasn't “healthy” and I had to read the YAML. Read the recent writing ↓
Clearing context in Claude Code throws away everything the session learned. I built a skill that writes a handoff document first, so the next session resumes instead of re-deriving. Install it with one command.
Every requirement I missed was one I was certain I'd implemented. The problem isn't carelessness, it's that self-review checks the code you wrote instead of the requirements you were given. I built an agent that inverts the order.
Microsoft ships an official Azure cost optimization skill for GitHub Copilot which is usable for any agent. Here is how it actually works in production, what it misses, and how to do the same thing without Copilot.
When Claude Code's context window fills up, output quality tanks: hallucinated imports, phantom files, broken logic. I fixed it by adding a statusline that shows context usage in real-time, and built a tool so you can do the same.
Green dashboards, running pods, low CPU, and yet production felt broken. A story about why Kubernetes metrics can lie about user experience, and what to watch instead.