re:shift Ricardo Lourenço engineering blog

NVIDIA GPU Passthrough on Fedora with dual vendor GPUs

Getting VFIO GPU passthrough working on Fedora 44 with an Intel Arc for display and an NVIDIA RTX 3060 Ti for VM compute.

Read article →

Latest Posts

GPU Cluster Networking on Kubernetes

An architectural deep-dive into how containers get direct GPU-to-network and GPU-to-storage access on Kubernetes with SR-IOV, Multus CNI, GPUDirect, and the full NVIDIA communication stack.


Why a 2B Model Beats an 8B Model: Evaluating LLMs Beyond Accuracy

Accuracy alone does not tell you if a model is useful in production. A composite scoring framework reveals that architectural design matters more than parameter count.


Three Experts, One Model: Making Small Language Models Reliable

How multi-perspective prompting and self-verification can make small language models produce reliable predictions without scaling to larger, more expensive models.

See all posts →

Projects

OpenShift Fake GPU Operator

A Kubernetes operator that simulates GPU resources on OpenShift clusters for testing and development purposes, enabling GPU workload testing without physical hardware.

Go Kubernetes OpenShift Operator SDK

Ansible Automation Workbench

Web-based UI for managing AWX and Ansible Automation Platform environments. Single Go binary with embedded React frontend.

Go React TypeScript Kubernetes Helm

rfplourenco.com

This personal site and blog, built with Astro and Tailwind CSS. Static site generation, Markdown content collections, dark mode, and RSS feed.

Astro Tailwind CSS TypeScript
See all projects →