DevOps · MLOps · AIOps platform engineer. I run Kubernetes for a fleet of robots, the CI and GPU platform behind it, and the ML training and inference pipelines on top, and I fix what breaks along the way in the open-source tools we depend on.
Golden Kubestronaut (every CNCF and Linux Foundation cloud-native certification)
Bugs found while operating these projects in production, fixed with a regression test, and merged upstream.
| Project | Change | Status |
|---|---|---|
| Kubernetes security-profiles-operator | operator and webhook run as non-root with all Linux capabilities dropped, with a render test that pins it | merged |
| cert-manager | certificate-shim no longer mutates the informer cache's labels; backported to 1.21 and 1.20 at the maintainers' request | merged |
| actions-runner-controller | default and per-controller max-concurrent-reconciles flags for the GitHub Actions runner controller |
merged |
| slam_toolbox | ROS 2 Humble backport of the save_map name validation that closed a shell-injection path |
merged |
| rclnodejs | interrupted ROS 2 message generation no longer leaves a partial generated/ tree that init() treats as complete |
merged |
| goreleaser | git tag parsing no longer breaks when the user's gitconfig sets column.ui |
merged |
Open work in review across Kubernetes, Argo CD, Kueue, Tempo, linkerd, Harbor, kops, aws-load-balancer-controller and others: all pull requests · issues filed
- MLOps on EKS: GPU batch and distributed training with Kueue, MultiKueue and Slurm on Kubernetes; Argo Workflows for data and inference pipelines; MLflow for experiment tracking; Karpenter spot capacity for GPUs; KubeRay pilots.
- CI platform: GitHub Actions runners on EKS with Karpenter, container build caches and registries, benchmark and GPU test suites.
- AIOps: LLM agents in the operations loop, from incident triage against Prometheus, Loki and Tempo to code changes that only land after mechanical checks.
More on Credly.