Study nature

Applied AI & Agentic Systems

What becomes possible when intelligence becomes cheap enough to apply at scale? Autonomous batch processing, agentic workflows, and problems that were economically intractable only a few years ago. I’m particularly interested in the emerging JavaScript and TypeScript AI ecosystem, from LLM harnesses and agent runtimes to RAG, embeddings, multimodal models, and large-scale deployments on Google Cloud.

Machine Learning & AI

I started a PhD in Pattern Recognition and Artificial Intelligence, where I chose to focus mostly on symbolic AI, but dropped out before completing it. A couple of decades later, I came back to find that I had backed the wrong horse. Meanwhile, transformers, self-supervised learning, foundation models, scaling laws, in-context learning, RLHF, multimodality, and inference-time reasoning had happened. There is quite a lot to catch up on.

Industrial Automation & Automatic Code Distribution

I've been experimenting with Python-based firmware for industrial automation, built around a custom UDP multicast discovery protocol. A particular interest is automatic code distribution: how networked controllers can discover, receive, and run software with minimal configuration.

High-Speed PCB Design

I've maintained an interest in high-speed PCB design and the less forgiving parts of electronics: EMI and noise mitigation, transmission lines and impedance mismatches, crosstalk, high-frequency return paths, board stacks, power/ground planes, and trace geometry.

Small-Scale Manufacturing on Demand

Obsessed with the D2C JLCPCB / SendCutSend / Xometry playbook: instant quotes, deep customization, no premium. None of the old circus of phone tag, sticker-shock quotes, and “come back when you hit our minimum.” This feels like a big change, and, in my opinion, it’s one of the ways we will save the American manufacturing base. See also The Zero-Inventory Hardware Company I.

Finite-Difference Time-Domain Simulators

Back in the day, I was extremely interested in numerical analysis techniques for modeling computational electrodynamics. I spent an unhealthy amount of time optimizing FDTD algorithms to run efficiently on the hardware available in the early 2000s.

Aren't they pretty? :)

Formal Languages & Compilation

I've had a long-running interest in formal languages, particularly regular expressions, grammars, parsers, and template compilers. It eventually turned into Kit55, a templating and static-site generation system based on Jinja2/Nunjucks.

Computer Graphics

Building EM propagation visualization tools first introduced me to computer graphics, initially through DJGPP and later Allegro, a library widely used by game programmers at the time. Since then, I've maintained an interest in the fundamentals of computer graphics, particularly the design of complex editors and visual user interfaces.