From software to hardware

This is a public record of my move from software and machine learning into hardware.

I have spent more than twenty years in ML, data, and software, across the full stack of it: data science, the data pipelines and warehouses that feed it, and running large-scale ML systems in production. None of it came from a degree. I am self-taught, and every real step forward has come the same way, by building something, watching it break, and fixing it. Learning by doing has always been my engine of growth, and it is exactly why I trust it to carry me into something new.

I looked at the usual options first: online courses, certificates, MOOCs, even Master's programs. They were too slow, too theoretical, and they did not let me build often enough. So I used AI to design my own one-year curriculum instead, a weekly loop of read, build, and experiment. You can read the whole plan. On the physical side I am close to a beginner; my hardware experience is a bit of Raspberry Pi home automation and a few small projects a while back. That is the appeal: to see how far first principles and a lot of hands-on building can carry a software person into making real, physical things, not just digital ones.

The name is not an accident. I bought granularity.ai about ten years ago and never knew what to do with it, beyond liking the idea itself. Granularity is just first principles applied, and it is how I actually work. Faced with anything complex, my instinct is to zoom in until I reach the smallest part that still means something, understand exactly how it behaves on its own, then step back and watch how it behaves next to everything else. Get the grains right, understand the seams where they meet, and the complicated thing mostly builds itself. In software this became second nature: unit tests for the parts, integration tests for the seams, and never trusting a system I could not explain one layer down. I expect hardware to reward the same discipline, only with less forgiving feedback. A wrong assumption in code throws an error; a wrong assumption in a power budget lets the smoke out.

I also like the word as a quiet counterpoint to "singularity." Singularity is the story where everything collapses into one sudden, unknowable leap. Granularity is the opposite bet: that real capability is earned from the bottom up, one well-understood detail at a time, and that if you respect the small stuff you can build almost anything.

So that is what this is: a public, granular account of learning to build physical systems, written as I go. The motto is the whole thesis: you can just build things. I do not need a certificate or a degree to start, only the fundamentals, the time, and the willingness to be bad at something new in public. If any of it grows into something commercially viable, even better, but that would be a happy side effect. The real objective is the learning.