What the work actually involves

You write and review evaluation items drawn from the deliverables you already produce: scientific slide decks, advisory board reports, medical education modules, HCP-facing web content. A typical task cycle is reading a model-generated asset — a slide narrative, a congress summary, a patient-versus-HCP adaptation — and deciding whether it holds together scientifically and structurally. That means checking whether a claim is supported by the cited trial, whether efficacy language carries the safety counterweight fair balance requires, whether an audience switch actually changed the register or just swapped vocabulary, and whether a graphic's logic survives the text it sits next to.

The deliverable is rarely a score on its own. It is structured written feedback: what the model got right, exactly where it drifted, and what the corrected version should have said. Engineers who do not have a medical affairs background read this output, so vague objections like "tone is off" are not usable. You will also be pulled into framework work — helping define rubrics and benchmarks for MedComms generation, arguing for where the pass line sits on things like promotional drift or over-simplification of a mechanism of action.

What the screen looks for

  • Depth of real deliverable experience — the AI interviewer probes for specific asset types you have owned end to end, and will follow up on details a reviewer-only contributor would not know.
  • Compliance reasoning, not compliance vocabulary — you can cite fair balance and off-label constraints, but more importantly explain how you resolved a specific dispute with a medical/legal/regulatory reviewer.
  • Source-tracing habits — how you verify a slide statement against the primary publication, and what you do when the citation is technically real but the data does not say what the slide claims.
  • Written critique quality — the ability to convert a judgment into feedback a non-clinician engineer can act on.

Logistics

Remote, contractor, asynchronous. Work is claimed from a queue in blocks; most contributors on projects of this shape run 10–20 hours a week with flexibility on when those hours fall, though calibration sessions and framework discussions may be scheduled live. Pay observed in the $50–80/hr range depending on therapeutic depth and whether you take on rubric and review work beyond straight task authoring. No prior AI or annotation experience is expected — the domain judgment is the asset being purchased.