Software that has to live close to the hardware — on boards, at the edge, under a power budget.
Cliqueraft is a small, independent setup — closer to a workbench than a company in the usual sense. The work sits at the edge of the network: embedded boards, sensors, and small robotics-adjacent hardware, running software that has nowhere to hide if it runs slow.
A large share of the time here goes into GPUs and the low-level end of machine learning — CUDA, quantized models, and inference squeezed onto hardware that was never really built for it. The interesting question isn't how large a model can get, but how much it can still do once almost everything has been taken away from it.
Power shows up as a real constraint, not an afterthought — boards that idle on a few hundred milliwatts, batteries that need to last months instead of hours. This page tries to follow the same rule: no web fonts, no trackers, nothing loaded that doesn't need to be, so it stays light wherever it's opened.
Alongside the hardware, Cliqueraft also builds a small number of niche Android apps — narrow, specific tools meant to do one job well and then get out of the way. They're kept alive quietly, the same way most of this work happens: over long stretches, in the background.
Over time, the direction has been drifting toward research questions in the same territory — embedded intelligence, low-power inference, that kind of thing — without any particular product waiting at the end of them. Nothing formal. Just where things seem to be heading.
This page isn't trying to sell you anything. If any of this overlaps with what you're working on, write in — [email protected].