Samuel Wibrow
Staff Engineer, AI Platforms · Zürich
Hexing YAML into compliance
~/github
~/cluster
~/recent-posts
view all →RTX PRO 4000 Blackwell vs RTX 3090 Ti for local LLMs: Qwen3.8-27B benchmarks
Is a single-slot RTX PRO 4000 Blackwell a viable second GPU instead of another 3090 Ti? llama.cpp benchmarks on Qwen3.8-27B, power cost in CHF, and both cards split together
Fine-tuning Gemma to write like me
QLoRA on Gemma 4 12B with 17k of my own Slack, email and WhatsApp messages, trained on a 3090 Ti in my home Kubernetes cluster
Sandboxing AI Agents for Garrison in Kubernetes
Why I stopped letting AI coding agents run against my home-ops cluster directly, and how kubernetes-sigs/agent-sandbox (SIG Apps) boxes them into pausable, disposable Kubernetes sandboxes instead
Verifying the IAM Role Attached to a Pod with EKS Pod Identity
Poking the Pod Identity Agent from inside a running pod to confirm which IAM role you actually got, with curl, wget, and a debug sidecar fallback for distroless containers
One alias to view or create a GitHub PR
A simple shell alias that saves a surprising amount of context switching.