Below the abstraction layer.

I started closer to the hardware.

During my computer engineering degree, I designed digital circuits and PCBs, described processor architectures in VHDL, wrote x86 assembly and built tools in C and C++. One early project was a compression utility; others were small Linux command-line programs. I liked having to understand every layer.

Those skills still come up in unexpected places: wiring encoders on a robot, chasing down a flaky camera connection, or debugging why an AI service runs differently on a server than on my own machine.

Things on my workbench.

I've assembled more than a few computers, soldered and rewired existing hardware, and worked with microcontrollers, sensors and small displays. At ko-br, I've wired encoders for a lead arm and spent a surprising amount of time proving whether camera dropouts were a hardware, USB or software problem.

At home there's usually a Raspberry Pi, an Arduino, or a Home Assistant experiment half under construction—screens, microphones, Sonoff devices, whatever makes an idea testable. I enjoy the build as much as the result.

Then making it all talk to each other.

I build my own machines for local model inference and run Linux servers with Docker, Open WebUI and reverse proxies. I've also run Ethernet cabling through houses and crimped RJ45 connectors. It's not the flashiest side of AI engineering, but a lot of the reliability comes from exactly this work.

Earlier, I built and maintained fleets of Raspberry Pi devices for remote content delivery, worked on backend software for electronics manufacturing, and took freelance jobs building APIs and Django applications. Those projects taught me a lot about supporting software after it's been deployed.