Agent runtimes & harnesses
How context, model calls, tools, permissions, memory, and feedback become one coherent execution system.
Interests / the questions behind the work
What keeps pulling me in.
I am most interested in systems whose behavior becomes clearer when you can trace the full loop.
Observable event-driven systems
Separating durable facts from live progress, preserving ordering, and making state transitions easy to inspect and replay.
Local-first AI tooling
Useful interfaces that keep sensitive logs, prompts, tokens, paths, and credentials under the operator’s control.
Learning through executable slices
Turning a complex architecture into one runnable path, then expanding it with tests and concrete evidence.
Linux desktop as a system
ArchLinux and the Wayland compositor stack (KDE, Hyprland, end-4 dotfiles) as a daily driver — configuration shaped into a maintained craft.