I'm Vib.

SplitStep

When I moved to LA I didn't know many people who played, so while I rebuilt my roster of hitting partners, solo practice had to actually count. I'd also spent years admiring broadcast tech like Hawk-Eye, and just as many watching amateur matches dissolve into line-call arguments nobody could settle, with no honest way to tell whether your game is even improving. The pro-grade answer costs more than most amateurs' rackets. So SplitStep: tennis training from phones you already own. No sensors, no subscription, no cloud in the sensing path. Two ordinary phones on court pair themselves over local Wi-Fi, sync their clocks to under a millisecond, and then make coordinated measurements with their mics and cameras together: the microphones establish exactly when a ball was struck, and which side it happened on, from the tiny difference in when each phone heard it, while the cameras track where the ball went. The vision half runs on a 1.16 MB ball-detection CNN I trained from scratch in PyTorch, taught first on synthetic rendered balls, graded only on real photos it had never seen, and improved until its accuracy more than doubled across five versions. A single training run exports the same weights to Android, iOS, and my research bench. Building systems that perform far above their cost, accessible to the everyday player instead of just the tour, is the same lean playbook that let CannMenus outrun better-funded rivals. Provisional patent drafted.