excloud
platform engineer
jun 2023 — aug 2026
remote
excloud was small enough that abstractions did not protect you for long. if a volume attached twice, a host disappeared halfway through provisioning, or a cluster stalled during bootstrap, the people who designed the path also had to make it recover.
i worked across most of that path. the interesting part was rarely creating the resource. it was deciding what should happen when reality and the database disagreed, then making retries safe enough that an operator could stop being afraid of them.
- compute / storage
- qemu/kvm control planes and spdk + nvme-of block storage, with reconciliation around provisioning, attach/detach, resize, cleanup, and recovery after interrupted transitions.
- clusters / network
- managed kubernetes from bootstrap through cilium, csi, and karpenter; identity, authoritative dns, and the arp/ndp machinery underneath the network.
- developer surface
- a shared go http layer that derives openapi 3.1 from types, then the sdk, cli, terraform provider, console, and telemetry around the services that use it.
4,000+ accounts380 active vms140 managed postgres clusters20+ internal services