I'm not an enterprising man, I'm an enterprise, man.
Platform engineering looks chaotic from the outside because it is — deliberately. The job is to absorb complexity so researchers and product engineers don't have to. Every weird half-finished repo here is a load-bearing wall in something larger.
I build the substrate: GPU compute platforms, inference gateways, agent scaffolding, GitOps pipelines. The things that make other things possible. If it looks like a junk drawer, you're reading the inside of the engine.
GPU compute-as-a-service on EKS — Karpenter autoprovisioning, EFA for multi-node training, FSx for Lustre with NVMe caching, SkyPilot interface, Flux GitOps. Built so ML researchers can submit a job without knowing what a NodePool is.