What the work actually is

Turing builds environments that test whether AI models can operate real engineering and scientific software the way a practitioner does. Your unit of work is a template: a structured record of one application's genuine capabilities — accepted input types, the operations that matter in practice, what it exports — plus the base assets a task runs against. That means writing Gaussian input and log files, OpenFOAM case dictionaries (`controlDict`, `fvSchemes`), WRF `namelist.input` decks, FITS images, SAC seismic traces and NetCDF climate datasets that a working colleague would accept as real. Internal consistency is the whole game: mesh, scheme, timestep and tolerance have to agree with one another, and a grid spacing that contradicts the physics option you selected is a defect even if the solver runs.

You also produce reference solutions — you complete tasks in your own domain on record, which proves the task is humanly achievable and fixes the standard for correctness. You are explicitly not writing scoring code. You define what correct means and why, in prose an engineer can turn into a check. The screening leans hard on that "why": a candidate who can say a boundary condition is right but not explain the physical reasoning behind it tends not to pass.

What the platform screens for

  • Five or more years working as a modelling and simulation physicist, with simulations you actually set up and interpreted — not software you were trained to click through.
  • Hands-on depth in at least one of Gaussian, OpenFOAM, SU2, Elmer FEM or WRF. This gate is real; expect follow-ups on specific files, flags and failure modes.
  • Fluency in differential equations, boundary conditions, wave propagation and solver convergence behaviour, including how you diagnose divergence rather than just restart with a smaller timestep.
  • Written English clear enough that a capability map or correctness definition needs no translation.
  • Helpful but not required: ObsPy, SeisComP, SAOImage DS9, CDO/NCO, and Python for generating input decks and post-processing output.

Logistics

Fully remote contractor assignment with flexible hours, but the stated commitment is 40 hours per week with at least four hours overlapping PST. The contract is scoped at five weeks with immediate onboarding — treat it as a short, dense engagement rather than an ongoing side project. No medical or paid leave. Pay is described only as "competitive"; no band was disclosed in the listing, so ask early and take any figure you hear elsewhere as observed rather than promised.