What the work involves

Most of your time goes to two activities. First, authoring evaluation items drawn from realistic scenarios: an FDA information request on a Module 2.7.4 safety summary, an EMA major objection on benefit-risk framing, a PMDA query about bridging data. Second, reviewing model-drafted responses and clinical documentation against those scenarios — deciding whether conclusions are actually substantiated by the cited data, whether commitments are scoped so a sponsor could keep them, and whether the record would survive a reviewer's second read.

A recurring theme is separating substance from polish. Language models produce regulatory prose that reads fluently and cites the right guidance titles while quietly asserting things the data do not support. Your job is to catch that, and to write a rationale grounded in a specific guideline, section, or precedent rather than "this feels thin." You'll also assess consistency across document sets — flagging where a claim in a clinical overview conflicts with the underlying summary, or where two responses in the same thread contradict each other.

What the platform screens for

  • Hands-on ownership of clinical submissions or health authority correspondence, not adjacent exposure. Expect follow-ups on specific filings you worked, your role in them, and the reasoning behind decisions you made.
  • Working fluency in the ICH framework and CTD/eCTD structure — where content lives, and why it lives there.
  • Practical knowledge of at least one major agency's expectations (FDA, EMA, MHRA, PMDA, Health Canada), including how that agency's queries typically read.
  • Written reasoning that is concise, defensible, and traceable to a source. Screens usually include a short written exercise or a live probe where you critique a flawed passage.

Logistics

Fully remote, contractor engagement, asynchronous. Observed pay for this role sits at $50–70/hr, varying with experience and specialization; rates are as reported and not guaranteed. Most contributors work 10–20 hours per week on their own schedule, with occasional synchronous calibration calls. No prior AI or machine learning experience is expected — the training is on the task format, not on your domain.