Blog
Notes on software, AI, and the industries we build for. Precise, without the hype.
- 2026-08-11
Ernst Klugmann Is Us: Translating Four Untranslated Latin Political Texts With LLMs
We used LLMs to translate four early-modern Latin political treatises into German — the first German translations of these texts ever — and we're self-publishing them under the pseudonym Ernst Klugmann. Here's how it went.
- 2026-06-02
Source-Anchored AI: Why Legislative Monitoring Can't Afford Hallucinations
General-purpose LLMs hallucinate, don't monitor proactively, and can't tell you what actually passed. Here's what source-anchored AI legislative monitoring requires instead.
- 2026-06-02
AI Works Better When It Has Less to Know
General-purpose AI tools are trained on everything and optimised for nothing. Vertical AI — built against a known, finite domain — can beat them on specific tasks, even with simpler architecture.
- 2026-05-22
Generating Lehrplan-aware lesson prompts
Why we feed a structured curriculum tree into the model instead of pasting Lehrplan PDFs into a prompt — and how that changes the output.
- 2026-05-10
A 60-line Go tool for validating .env files
A small utility we built to stop catching missing env vars in production. One file, one binary, no dependencies.
- 2026-04-28
Why we picked Hugo for this site
Static, fast, boring, and out of our way. A short note on tooling choice and the design system we ended up with.