TIL: Quarto, Small Language Models, Self-Hosting
TL;DR
Use Quarto for presentations and documents; If you’re hardware allows it, run small language models locally using Ollama. Give the latest iteration of Gemma a try; Seriously consider self-hosting if you have basic Linux and DevOps experience. A Linux server, reverse-proxy and container runtime is all you need to get started.
Quarto
Experimenting with Quarto has been on my list for a long time. This month I finally got to use it to create revealjs powered slides and a typst based document. I authored both documents in VSCode using the official Quarto extension and won’t go back.
Small Language Models
With the release of Gemma 4, I’ve updated my Ollama installation and took several small, open-source language models (Granite 4, Qwen 3.5, Gemma 4 and SmolLM v3) for a ride. My current hardware setup (MacBook Air M2 with 16 RAM) only allows me to run the tiny versions of those models. While all of them work great for prototyping AI-powered applications, they don’t yet replace frontier models for more advanced tasks. I might just need to invest in beefier hardware to make that happen.
Selfhosting
I went down the self-hosting and homelab rabbit hole after reading Michael Kennedy’s Talk Python in Production book. After researching self-hosted PaaS solutions, I tinkered with disco on a super cheap Hetzner server to host apps over at byteCare. Currently, I am implementing a custom solution based on Caddy, Docker Compose and Zitadel. Stay tuned, this might deserve its own blog post.