Engineering

4 AI models, a hybrid search engine, custom Metal shaders, and a notch HUD — all swapping in and out of 8 GB of unified memory.

~2,000 lines of Swift, two critical bug fixes, and an honest benchmark showing the off-the-shelf model beats our finetune.

We run two Whisper models simultaneously — one for real-time preview, one for accuracy. Here's the memory choreography that makes it work on 8 GB.

Bézier shoulders, spring physics, and one SwiftUI trick that makes a notch HUD feel like Apple built it.