Hello, I'm
Software engineer. 14 years of native iOS and tvOS work, now spending much of my free time on prompt engineering and AI evaluation.
Across 3,300 LLM outputs, I measured three distinct layers of variation. Models rephrase freely while recycling the same concepts and arguments. Temperature doesn't help. A few structural prompt techniques do.
I built a framework that places LLM agents in constrained creative environments with synthetic backstories and emotional memories, then measured whether it actually makes them more creative.
First post — what this site is about and what's coming.