What the work actually is

You are not labelling data or rating chatbot replies. You author original research problems — take a published paper, a Kaggle dataset, an open-source bioinformatics repo, or a scenario you design, and turn it into an executable task with a defensible ground truth. That means writing the scientific prompt, building the input files or environment the model needs, and writing grading criteria precise enough that a correct-but-differently-structured answer still passes and a plausible-looking wrong answer still fails. Then you calibrate: run it against frontier models, and if they solve it comfortably, it does not ship. Tasks that survive are usually the ones where the biology is subtle — a normalisation step that changes the differential-expression call, an ecological model whose assumptions break on the given data, a kinetics fit that only works if you handle the enzyme's substrate inhibition.

What the screen looks for

The 25-minute conversational interview covers background, experience and motivation, but the substance is domain depth under follow-up. Expect to be pushed past your first answer: name a subdomain, then be asked what specifically you did in it, what tooling, what went wrong. Mercor asks for demonstrated depth in at least two of ecology, biochemistry and genetics, with a coding focus — so a genetics PhD who has never touched an ecological dataset should be honest about that and lean on the second subdomain they genuinely have. The engineering side is a real filter, not a nice-to-have: authoring runs through pull requests with automated quality checks and Docker execution, so if you have only ever run scripts in a local notebook, say so and say how quickly you can close the gap.

Logistics

  • Fully remote and largely asynchronous; you set your own hours within the 20+ per week commitment.
  • Six-week engagement with an immediate start — this suits people with existing bandwidth, not people finishing a thesis in eight weeks.
  • Observed pay for this listing is around $70/hour; rates on Mercor vary by domain, seniority and project, and are set per engagement rather than guaranteed.
  • Expect review cycles: a task can come back for tightening of the rubric or the environment before it merges.