Building a cloud platform where everything is just another Kubernetes resource
We turned the Kubernetes API into the control plane for a whole cloud platform, not just for containers. S3 buckets, databases, DNS records, IAM credentials and even child clusters are YAML manifests, managed with GitOps and reconciled by controllers. We started on Terraform, and it held up until monolithic state files started locking the whole team, applies recalculated everything to change one thing, and drift only showed up when someone remembered to run a plan. This is the rebuild on Crossplane and ArgoCD: the production architecture, what broke during the migration, and what I would do differently now.
Watch on YouTube →What makes a good tech team (keynote)
Software teams that fail rarely fail on talent. They fail on how they work: code gets discussed instead of tested, formatting gets debated instead of automated, ego gets optimised instead of outcomes, and nobody ever asks whether any of it works. Drawing on Google's Project Aristotle, the DORA metrics from Accelerate and Team Topologies, this keynote walks through eight patterns that separate teams that ship from teams that stall. You leave with a checklist for your next interview to work out whether a team is worth joining. Held in German.
Watch on YouTube →Unveiling the magic: scaling large language models to serve millions
Ever wondered what sits behind the curtain when you chat with something like ChatGPT? This talk goes through the challenges of serving large language models at scale: acquiring and caching models, building an OpenAI-compatible server, authenticating requests, tracking usage for billing, autoscaling, and hosting several models under one domain.
Watch on YouTube →Insights into managed service provision: a STACKIT retrospective
Building a cloud service is more approachable than it looks from the outside. This retrospective covers how we built and ran managed services at STACKIT, and what the work actually involves once real customers depend on it.
Watch on YouTube →Rust beyond systems: revolutionizing web development
Rust gets filed under systems programming, but it is a general-purpose language: Tauri for desktop apps, Leptos on the frontend, Axum on the backend. I learned Rust by building a web application, which as a backend engineer was the ground I knew best, and found the language fits web work well. The talk covers why its guarantees give me confidence in the applications I ship.
Watch on YouTube →Everyone loves talking about architecture. Few mean code architecture
Everyone either wants to be an architect or already claims the title, across every domain: data architects worrying about governance and lineage, customer success architects making sure clients hit their goals. This talk brings it back to the code, and how to structure it so a system welcomes new people, survives a few years, and holds up against bugs. Held in German.
Watch on YouTube →Building resilient distributed systems: lessons from practice
Distributed systems draw equal parts awe and dread. Message queues, homegrown microservices, other teams' services and third-party dependencies all have to hold together. This talk is about what makes such a system survive contact with production. Held in German.
Watch on YouTube →