What the work actually involves

Turing builds environments that test whether AI models can operate real desktop engineering and scientific software. This role sits upstream of task generation: you build the template for a domain — a structured record of what an application supports (input types, the operations that matter, export formats) and the realistic material a task is later constructed from. Day to day that means writing Octave scripts and GeoGebra worksheets that a working numerical analyst would recognise, assembling numerical dataset CSVs and target matrix outputs whose numbers actually agree with each other, drafting mathematical proof specifications, and then solving your own tasks on record so there is evidence the work is humanly achievable.

You are not writing scoring code. You define what correct means — tolerance on a convergence result, which intermediate quantities must match, what counts as an equivalent construction — and engineering implements the checks. That division makes the written articulation of correctness the core deliverable, not the artefact alone.

What the screen looks for

  • Authored algorithms, not scripted workflows. The listing is explicit: five-plus years in scientific computing, applied mathematics, or statistical modelling, where you have written real computational algorithms rather than run packaged ones.
  • Hands-on time in Octave or GeoGebra. At least one is mandatory. Expect follow-ups on specific functions, file formats, and the boundaries of what the tool can and cannot do.
  • Why, not just that. Grounding in numerical linear algebra, ODE solvers, regression modelling, and hypothesis testing — assessed by whether you can justify a convergence result or a geometric construction, not only produce one.
  • Internal consistency discipline. Base assets are graded on whether details agree across files; a CSV whose summary statistics contradict the script that produced it fails.

Logistics

Fully remote contractor assignment, 40 hours per week with at least four hours of overlap with PST. The contract is stated as five weeks with immediate onboarding; extension is not promised. No medical or paid leave. Pay is undisclosed on this posting — Turing describes it as competitive, and rates in comparable expert-data work vary widely by domain and seniority, so treat any figure you hear as observed rather than guaranteed and confirm it before signing.