Kubernetes & Cloud Native Berlin Meetup - August Edition
SCHEDULE:
[17:00 - 18:00] Doors open - Get a juice/soft drink, relax
[18:00 - 19:00] Dr. Stefan Schimanski & Sergiusz Urbaniak - From Von Neumann to Transformers: An Engineer's Guide to What's Actually Inside an LLM
[19:00 - 19:10] Q&A
[19:10 - 19:45] Break - time for networking with pizza and other refreshments
[19:45 - 20:15] Christoph Voigt - Building a Kubernetes Cluster API Provider with Agentic AI: What Actually Changed?
[20:15 - 20:25] Q&A round
[20:25 - 21:00] More time to network with peers and colleagues from the industry at a relaxed pace
TALK DETAILS:
From Von Neumann to Transformers: An Engineer's Guide to What's Actually Inside an LLM
For decades, we've understood computers through the same mental model: instructions, registers, caches, memory hierarchies, and CPUs executing programs one step at a time. That intuition scales remarkably well—from an 8086 diagram to a modern Linux kernel running Kubernetes.
Then LLMs happened.
Today we use them to write code, build agents, and solve increasingly complex problems. Yet the model itself often remains a black box: prompts go in, tokens come out, and everything in between feels like magic.
This talk opens that box.
Starting from a transformer trained from scratch, we'll build a mental model for how an LLM actually computes. We'll explore tokenization, embeddings, attention, feed-forward layers, and the flow of information through the network, showing how these building blocks form a new kind of compute architecture. Finally, we'll follow backpropagation as it gradually reshapes millions of parameters, turning raw text into a model capable of understanding language and reasoning about code.
No machine learning background required—just curiosity and the systems intuition you've already built.
Building a Kubernetes Cluster API Provider with Agentic AI: What Actually Changed?
Can a modern AI agent implement a non-trivial Kubernetes project?
To find out, I built a complete Cluster API provider for STACKIT almost entirely through agentic software development.
Cluster API turns out to be a surprisingly good benchmark for AI. Providers are largely defined by APIs, contracts, reconciliation loops, and Kubernetes conventions. There is little room for creativity but a lot of room for correctness.
In this talk, I'll share what worked, what didn't, and why the biggest challenge wasn't generating code it was validating it. We'll look at practical lessons from building a real Kubernetes project with AI and discuss what this experience might tell us about the future role of software engineers.
Whether you're interested in Kubernetes, platform engineering, or AI-assisted development, expect an honest engineering case study rather than another AI hype talk.
-
6:00 PM - 7:00 PM CEST
From Von Neumann to Transformers: An Engineer's Guide to What's Actually Inside an LLM
in-personFor decades, we've understood computers through the same mental model: instructions, registers, caches, memory hierarchies, and CPUs executing programs one step at a time. That intuition scales remarkably well—from an 8086 diagram to a modern Linux kernel running Kubernetes.
Then LLMs happened.
Today we use them to write code, build agents, and solve increasingly complex problems. Yet the model itself often remains a black box: prompts go in, tokens come out, and everything in between feels like magic.
This talk opens that box.
Starting from a transformer trained from scratch, we'll build a mental model for how an LLM actually computes. We'll explore tokenization, embeddings, attention, feed-forward layers, and the flow of information through the network, showing how these building blocks form a new kind of compute architecture. Finally, we'll follow backpropagation as it gradually reshapes millions of parameters, turning raw text into a model capable of understanding language and reasoning about code.
No machine learning background required—just curiosity and the systems intuition you've already built.
SPEAKERS -
7:45 PM - 8:15 PM CEST
Building a Kubernetes Cluster API Provider with Agentic AI: What Actually Changed?
in-personCan a modern AI agent implement a non-trivial Kubernetes project?
To find out, I built a complete Cluster API provider for STACKIT almost entirely through agentic software development.
Cluster API turns out to be a surprisingly good benchmark for AI. Providers are largely defined by APIs, contracts, reconciliation loops, and Kubernetes conventions. There is little room for creativity but a lot of room for correctness.
In this talk, I'll share what worked, what didn't, and why the biggest challenge wasn't generating code it was validating it. We'll look at practical lessons from building a real Kubernetes project with AI and discuss what this experience might tell us about the future role of software engineers.
Whether you're interested in Kubernetes, platform engineering, or AI-assisted development, expect an honest engineering case study rather than another AI hype talk.
SPEAKERS