I learn by building — most of what I know came from picking a real problem and shipping something for it, not from courses. I'm drawn to developer tools and ML: the kind of work where you can feel whether something is actually solid or just looks like it. :P
A deterministic execution sandbox and competitive-programming judge, built to enforce hardware-level CPU and memory limits with microsecond accuracy. The core is a low-overhead Rust engine using POSIX primitives (fork, execve, setrlimit) with wait4 kernel telemetry, backed by a zero-config subprocess fallback for platforms without it.
A self-organizing, adaptive list — a small model that learns to resurface items by usefulness rather than relying on fixed sort order, aimed at the kind of unstructured working lists AI engineers accumulate. The blog above covers the approach and the reasoning behind it.
Elsewhere