What the work actually is

Mercor is building a scientific computing benchmark (Sci Code) with frontier AI labs, and this track covers chemistry. You bring your own source material — a published paper, an open dataset, a repo you know well, or a scenario you construct — and turn it into an executable research problem with a defensible ground truth. That means writing the prompt, writing the reference solution, and writing the grader: the explicit criteria that separate a correct answer from a plausible-looking wrong one. A task only ships if strong models fail it more often than they pass it, so calibration against frontier models is part of authoring, not a separate QA step.

Expect the chemistry to be real: SCF convergence behaviour, basis set and functional choices, post-Hartree-Fock method selection, PES scans, thermochemical corrections, MD setup and analysis, cheminformatics pipelines. Expect the engineering to be real too — tasks run in Docker, submissions go through pull requests, and automated quality checks will reject work that isn't reproducible. If your code has only ever run on your own cluster with your own module environment, budget time for that adjustment.

What the screen looks for

  • Verifiable depth in two subdomains. The listing asks specifically for quantum chemistry and computational chemistry. A screener will follow up on specifics — methods you've actually run, packages you've actually used, where they broke.
  • Grading judgment. Can you state what makes an answer wrong when the number is close? Numerical tolerance, unit handling, convention differences (e.g. sign, reference state, basis set superposition error) are where graders fail.
  • Adversarial instinct. Can you describe a problem a strong model would confidently botch, and say why — not just "it's hard."
  • Toolchain fluency. Git, PRs, Docker, deterministic runs.

Logistics

Remote and largely asynchronous. Six weeks, part-time, 20+ hours per week, immediate start. The application is a resume plus form, then a ~25-minute conversational interview on background, experience, and motivation, with next steps within a few days. Publications and prior research-engineering or scientific-software work are preferred, not required. $70/hr is the observed rate for this listing, not a guarantee.