Storia: How I use LLMs for structured classification without getting garbage output — Warptech Lab News