What the work actually involves

Each task you produce is a small, self-contained mathematics problem set: one main problem plus at least three sub-problems that build logically toward it. You write the specification so that it is well-posed — explicit constraints, defined inputs, unambiguous expected outputs — then implement a golden solution in Python with unit tests covering every sub-problem. The harder half of the job is test design: cases must be discriminative, meaning a model that reasons almost correctly still fails them, and deterministic, meaning no floating-point drift, no random seeds, no environment-dependent behaviour. Expect heavy use of NumPy, SciPy and SymPy, and expect to think carefully about tolerances.

Submissions go through Turing's Central Task Platform: Tier 1 structural checks, then Tier 2 quality rubrics, then Pass@K evaluation against multiple LLM judges (GPT, Gemini, Nemotron). Tasks that every judge solves on the first attempt are not useful; tasks that are unsolvable because the statement was ambiguous are rejected. You will iterate on QC feedback, and your standing on the project depends on rework rate — the target is consistent L1 approval on first submission.

What the screen looks for

  • A completed or in-progress Master's or PhD in Mathematics, and the ability to talk about your own research area with specificity.
  • Python that is genuinely production-adjacent, not notebook-only: you should be comfortable writing pytest-style assertions and reasoning about numerical determinism.
  • Judgment about difficulty calibration — can you explain why a given problem would or would not discriminate between frontier models?
  • Prior exposure to AI data work, coding benchmarks, or rigorous scientific writing is treated as a strong signal rather than a formality.

Logistics

Fully remote, but not fully async: the commitment is 40 hours per week with a four-hour daily overlap with PST, and you are expected on sync calls for reviews, feedback sessions and standups. The engagement is a contractor assignment of roughly eight weeks with no paid or medical leave. Hiring is restricted to Bangladesh, Brazil, Colombia, Egypt, Ghana, India, Pakistan, Indonesia, Kenya, Nigeria, Turkey and Vietnam. Pay was not disclosed in this listing; rates on Turing's STEM projects are typically set per-hour by project and seniority, so confirm the band in writing before you commit the hours.