Small, fast, portable, good. Pick four.
It started with a stubborn question: why can't search see meaning as you type, the way it already matches letters? One experiment led to the next, and this is where they landed.
The bird. An Arctic tern weighs about 100 grams and flies pole to pole every year: more than 70,000 km, the longest migration of any animal, on a body you could hold in one hand. It carries what it needs and goes an absurd distance on very little. That's the spirit here: small, self-contained, and still does a lot.
The questions. How far can you push the tradeoffs: small, fast, portable, and still ship something people use? And in JavaScript land: where's the Fuse.js of semantic search, something you just npm install and drop in? A couple of conversations later (one with an AI-engineer colleague, one with a frontend colleague), it had turned into a real idea.
Me. I'm an AI engineer, started in computer architecture, then C/C++, malloc and free and segfaulting my way through the early years. I loved working close to the metal: near line-speed, and all the OS-level control that comes with it. From there it was up the stack, one layer at a time: drivers, applications, UIs, and every flavor of build hell along the way (toolchain vs. target platform vs. the DLLs in between; later, the special Python packaging kind). The pull toward more general-purpose solutions eventually landed me in ML, training my own models. Tech generalist, AI specialist, curious above all.
The click. When I read the BitNet b1.58 paper, my head lit up. Here was the missing piece for a project that would touch every layer I'd ever worked on, top to bottom: not just something to build, but a way to express everything I'd picked up along the way.