Polaris was a confidential project. NDA prevents disclosing visuals or product specifics. What follows is the role, the working model, and what the work taught me, without revealing anything covered by the agreement.
What I did
A dual role at Apple: UX/UI design and front-end engineering held by one person. Vue.js + SASS on the implementation side. The design language aligned with Apple's Marcom style guide for visual consistency across surfaces.
How the team worked
The teams I worked with at Apple held design and engineering as a single craft. Not "design hands off specs to engineering." Design and engineering sat in the same standups, read each other's pull requests, jointly owned the production output. Specialty existed (you were either a designer or an engineer at the title level) but the practice didn't observe the line.
That's an unusual operating model and it produces an unusual quality bar. Decisions that would normally bounce between teams (should this animation respect reduced-motion? does this layout work at compressed pixel densities?) got resolved in real time because everyone in the room was qualified to weigh in.
What the work demanded
A higher floor for craft than I'd worked under before. Apple ships a lot of surfaces; the visible ones get scrutiny most product teams never see, and the internal ones inherit the same standards. Working in that environment recalibrates what "done" means. A pattern that would ship cleanly elsewhere wouldn't survive review at Apple, and the reasons, once you understood them, were almost always right.
What I took away
Vue.js was new to me at the time and the project deepened my front-end implementation work substantially. More importantly: the working model (design and engineering as one craft) became how I work everywhere since. Spellbook, Elsa, MixShift, this site itself. The integration isn't a process choice; it's a quality choice.
Why this matters for AI systems
The closest analogue to Apple's design-engineering integration that I've seen, in a different domain, is what high-functioning AI teams need. When the model researcher, the tooling engineer, and the product designer aren't speaking the same language, when handoffs happen between adjacent disciplines that don't read each other's work, AI systems ship late, ship broken, or don't ship at all.
The teams that ship production AI reliably blur those lines the same way Polaris blurred design and engineering. One craft, three specialties. Everyone reads everyone else's work.