ABOUT

My name is Daniel Cavalli. I'm a Machine Learning Engineer who studied economics, which, at the very least, gave me a formal excuse to be suspicious of systems that appear to work before I understand why they work.

Not that I needed formal education for that. That instinct has been fairly consistent in my life, and even as a child I would constantly at times get in trouble for taking apart something that was "broken", to better understand how it works.

Well, having formal education meant applying this same mentality to other, uhm, less concrete, topics. Sometimes it leads to distributed training stacks being repurposed, CUDA kernels being written for pipelines that were working, but barely. Other times it leads to surfboards being built in the storage room of my kitchen (with all the fiberglass, tools and dust that comes with it) or a washing machine that had been operating perfectly well, until I decided it should now contribute to science!

The object changes, but the interest is usually the same: finding the real logic underneath things, where the friction actually is, which parts are carrying their weight, and what can be made simpler in order to make it better.

My family, while supportive of this behavior, still gives me the occasional look here and there. Unwavering in my intent to break things in order to learn, this blog is where that habit becomes easier to justify. I write about artificial intelligence, systems design, networking and whatever else seems worth understanding properly rather than merely using. Partly to explain things and share some knowledge or opinions, but mostly to see whether my own explanations survive contact with structure and opposing views. Proud to say that they usually don't, which is useful.

I live in Copacabana, stay as close to the ocean as ordinary life allows, and have found that surfing, engineering, and writing are similar in at least one respect: all three are less impressed by intention than by timing, feel, and whether you actually understand the medium you're working with.