Posts
All the articles I've posted.
-
The Frikadeller Trial: How AI Judges Art
I put Claude Opus 4.7 on trial against 4.6 using the Delphi method and a three-judge LLM panel, before migrating production. Here is what the protocol found.
-
Intake Is the Architecture
Four intake pipelines feed one search layer on a $200 gaming laptop, turning Karpathy's LLM Wiki into a personal knowledge system that already knows you.
-
Turning on the lights
OTEL observability on Claude Code sessions revealed 277,000 wasted tokens in one workflow. The grep-before-read fix took fifteen minutes.
-
So the Recipes Have Soul
I built an autonomous AI research department to produce deep cultural research on culinary pioneers. Here's how it works and what it found.