System Design — Head of Software Engineering [GPU]

The big design walkthroughs for the role. Each is the kind of open-ended "design X" question a senior GPU-platform loop asks — answered at the altitude a Head of Engineering operates at: requirements → architecture → the hard tradeoffs → the failure modes → the commercial framing.

How to use these

1. Drive the requirements first — every good design answer starts by establishing scale, SLOs, tenancy, and constraints. The walkthroughs model this.
2. State the tradeoffs explicitly — the senior signal is naming what you're giving up (isolation vs utilization, abstraction vs performance, cost vs guarantee), not presenting one "right" answer.
3. Name the failure modes — at this level, "what breaks and how you detect/ recover" (Phases 08, 10) is half the answer.
4. Close with the commercial framing — a Head of Engineering connects the architecture to adoption/revenue (Phase 12 Ch. 6).

The design-interview rubric for this role

• Requirements & scoping: did you establish scale, SLOs, tenancy, hardware, sovereignty?
• Architecture: clean abstraction boundaries (the seams — Phase 12 Ch. 8); the right components.
• Depth on demand: can you go deep on any box (KV cache, scheduler, HAL ABI)? The phases are that depth.
• Tradeoffs: named and reasoned, not hand-waved.
• Operations: observability, release, failure (Phases 08, 10).
• Commercial judgment: the business framing (Phase 12).