>about

I'm Leon Lehmann — usually just Leon, and online as kodee. I'm the founder/CEO & researcher at Empero, where we work on making language models more efficient: better memory layouts, smarter residual and recurrence schemes, and honest evaluations of the exotic stuff.

research

Recent write-ups are on my alphaXiv author trail — recurrence doesn't pay, multi-head gated value residuals do, and n-gram memory is cheaper than you think. Links are on the home page. Code and weights ship under github.com/empero-org and huggingface.co/empero-ai.

colophon

This site is hand-rolled — a single stdlib-only Python server, no framework, no trackers, no cookies (the theme toggle uses localStorage, that's it). The header is a starry night generated and dithered at runtime in a canvas with a 4×4 Bayer matrix, tinted to the site's violet palette. Type is Space Mono and Space Grotesk.

contact

Fastest: x @kodeefr, or just email lehmann@empero.org. Also on linkedin and github. Or just sign the guestbook.