Architecture + scaling + technical leadership for AI systems.
Advanced transformer internals: MoE, KV cache, speculative decoding, scaling laws.
By the end of this week, you can argue MoE routing, KV-cache memory math, and scaling laws on a whiteboard with concrete numbers.
Staff rounds expect you to reason about MoE routing, KV-cache memory, and scaling laws without notes. This is the technical bar.
Watch out: Reading the GPT paper isn't enough. Staff candidates are expected to know the inference-time memory cost of attention by heart.
End to end LLM platform design: multi-tenant serving, cost optimization, reliability, monitoring.
By the end of this week, you can design a multi-tenant LLM platform with cost attribution and reliability targets.
Multi-tenant serving and cost-attribution questions are where staff candidates earn their offer. Bring numbers, not adjectives.
Watch out: Vague 'add a load balancer' answers don't land. Bring numbers about QPS, latency tails, and cost per request.
Alignment, safety, evaluation at scale, team structure for AI orgs, build-vs-buy decisions.
By the end of this week, you have a clear, defensible opinion on alignment, build-vs-buy, and AI team structure.
Staff hires need a clear opinion on alignment, build-versus-buy, and team shape. Hedging here reads as junior.
Watch out: Hedging is the staff-interview kiss of death. Have a strong opinion you can change with new evidence.