What the work actually looks like

You spend most of your day reading engineering problems and the model's attempts at them, then deciding whether the answer would survive a real design review. That means catching the things a well-written wrong answer hides: an assumed boundary condition that does not exist, a missing load case, a unit slip between lbf and lbm, a tolerance stack that cannot be held, a material choice that fails at service temperature, a factor of safety quoted without a governing code behind it. The other half of the job is generative — writing instruction specs that define what correct means for a task, producing worked reference solutions with the arithmetic and assumptions shown, and proposing new tasks that reflect how engineering is really done rather than how textbooks present it.

Calibration is a standing part of the role. You will be in written discussion with client researchers and with specialists in adjacent disciplines, pushing tacit judgment into explicit criteria that someone else can apply the same way you would. Precise written feedback matters more here than polish in a slide deck.

What the screen looks for

  • Ownership, not exposure. The platform probes for designs, analyses, and drawings you personally released — part numbers, load cases, test results, production handoffs.
  • Depth under follow-up in at least one subdomain: mechanical design, thermal/fluid, structural/stress, manufacturing, or controls. Expect the interviewer to keep asking why after your first answer.
  • Standards fluency you actually use — GD&T per ASME Y14.5, ASME BPVC sections, ASTM test methods, ISO tolerancing — rather than acronyms recited.
  • Evidence you have used LLMs on real engineering tasks and can articulate where they fail.

Logistics

Remote within the United States, W-2 through Cincinnatus LLC as employer of record, with the possibility of placement inside a leading AI lab's extended workforce. Full-time at 40 hours per week; client-issued equipment and accounts, and work happens inside the client's tools, so this is not an async side project. Pay has been observed in the $60–90/hr range and is not guaranteed; the initial engagement is 2–3 months with extension tied to performance and project needs.