What the work actually is

You write scientific coding tasks, not prose. Each submission is a main problem plus at least three sub-problems that chain logically toward the final result — for a chemistry author that might be building up from a partition-function calculation to an equilibrium constant, or from a rate-law fit to a full kinetic simulation. For every step you implement a verified golden solution in Python and write unit tests that actually discriminate: tests that a plausible-but-wrong model output fails, not tests that pass on any code returning a float. Determinism matters — seeded randomness, fixed tolerances, no floating-point comparisons that flip between runs.

Submitted tasks go through QC on Turing's Central Task Platform: Tier 1 structural checks, then Tier 2 quality rubrics covering well-posedness, scientific correctness, and test discriminativeness. Tasks are additionally run against multiple LLM judges (GPT, Gemini, Nemotron) under Pass@K criteria — a task that every frontier model solves on the first try is generally too easy, and one that is unsolvable because it is ambiguous rather than hard is a rework. You iterate on QC feedback, and the project tracks rework rate, so first-pass L1 approval is the standard being measured.

What the screen looks for

  • A Master's or PhD in chemistry or an adjacent field, and enough depth to construct a problem that is scientifically airtight rather than merely plausible-sounding.
  • Real scientific Python: NumPy, SciPy, SymPy, and whatever domain tooling you use (RDKit, ASE, PySCF, cclib and similar are all fair to cite).
  • Evidence you can specify a problem precisely — inputs, units, constraints, tolerances, expected output shape — because ambiguity is the single most common rework cause.
  • Prior exposure to annotation work, benchmark construction, or LLM evaluation is a stated plus and comes up in follow-ups.

Logistics

Remote, contractor/freelancer terms with no paid or medical leave, an 8-week contract, and a stated expectation of 40 hours per week with at least 4 hours overlapping PST — this is not a nights-and-weekends side engagement. Hiring is restricted to Bangladesh, Brazil, Colombia, Egypt, Ghana, India, Pakistan, Indonesia, Kenya, Nigeria, Turkey, and Vietnam. There are sync calls for reviews, feedback, and standups inside the overlap window. Pay is not disclosed in this listing; Turing typically quotes an hourly rate at offer stage and rates vary by country and credential.