Get a reply from this employer
Your resume and cover letter, tailored to exactly what they’re hiring for.
Turing is hiring contract engineers to author and score aerospace and flight-dynamics tasks used to train frontier AI models. You will work with OpenVSP, XFLR5, FlightGear+JSBSim, OpenRocket and QGroundControl, requiring genuine proficiency in at least two tools plus scripting.
The role emphasizes end-to-end task definition, reference solution creation, and rigorous scoring. You will review colleagues' tasks for technical correctness and completeness, ensuring assumptions are explicit and checks
Turing is hiring contract engineers to author and score aerospace and flight-dynamics tasks used to train frontier AI models, building golden geometries and mission plans in OpenVSP, XFLR5, FlightGear+JSBSim, OpenRocket and QGroundControl. The posting is filed under Electrical Engineering but the stated scope is aero/flight dynamics — expect the screen to test that domain specifically.
The tool list is specific: OpenVSP, XFLR5, FlightGear+JSBSim, OpenRocket, QGroundControl. Turing asks for genuine proficiency in at least two, plus scripting — parameterising geometry sweeps, batch-running cases, parsing output rather than reading it off a GUI. Note that the listing carries an Electrical Engineering title while the entire body describes aerospace and flight dynamics work; treat the body as authoritative and prepare accordingly.
You write engineering problems that a strong model should be able to solve and often cannot. That means defining a task end to end: inputs, assumptions, constraints, operating conditions, output metrics, and acceptance criteria that a grader can apply without re-deriving your intent. You then build the reference solution — a golden geometry, a mission plan, an airframe or route file — and score model-generated submissions against it. A meaningful share of the day is reviewing other engineers' tasks for technical correctness and completeness, which is where most quality problems get caught.
Turing's screen is AI-led and follow-up heavy. It probes whether you can state assumptions explicitly, whether your sanity checks are real (order-of-magnitude, conservation, known-solution comparison) rather than "it looked right," and whether you can distinguish a model answer that is wrong from one that is merely different from yours. Expect to be asked what a particular tool cannot do — vortex-lattice limits, JSBSim aero-table fidelity, OpenRocket's assumptions — because naming a limitation is the cheapest proof of hands-on use.