We're assembling a group of senior software and ML engineers to work on one of the hardest open problems in AI today: how well can frontier models reason about LLM inference systems, GPU-level performance optimization, and model-serving architecture? You'll help design evaluation scenarios, write reference solutions, and grade model outputs on real systems-engineering problems — the same class of problems you likely work on day-to-day. This is a research-and-evaluation role, not a traditional engineering job — you won't be shipping production code for us, you'll be defining what "correct" and "excellent" look like for AI models tackling the problems you already know deeply.