Cloud Native Frankfurt Meetup - September 2026

Turning a SAP operations team into a platform team that offers self-service via an Internal Developer Portal sounds straightforward on paper: build the portal, onboard the services, migrate the accounts, done. In practice, it's a different story. Over the past two years, I worked with an internal operations team at a KRITIS company to help them offer their services through an Internal Developer Portal and migrate all accounts onto it. The bigger bottleneck turned out to be something I hadn't put on the project plan: the team had little prior experience with Git, Terraform, or IaC-based self-service workflows. Like most platform engineers, I had simply assumed these skills would already be there. By the time pair programming sessions revealed that basic Git usage was still a daily struggle, our ambitious goals were already out of reach. This talk is an honest retrospective — not a blame game, but a close look at what I missed in planning, what I'd watch for earlier next time, and how I now think differently about what "platform readiness" actually means.

Running LLM inference on Kubernetes usually means stitching together a serving engine, GPU scheduling, model storage, and routing before you get a single token back. Modelplane is an open-source, Kubernetes-native control plane that turns that into a handful of APIs.

This is a hands-on walkthrough: we will go through the Modelplane API surface, what the resources are and how they fit together and then deploy a model end to end and send it a request. Along the way I will point out where the interesting problems live (GPU scheduling, model caching, routing) and what the project is doing about them.

No prior inference experience needed ;) basic Kubernetes familiarity helps