Mercor is recruiting radiologists, imaging fellows, and PACS-fluent technologists who use Visage 7 weekly to evaluate AI model outputs on real imaging-workflow tasks. Work is remote and largely asynchronous, with observed rates of $100–130/hr depending on credentials and depth of Visage 7 usage.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

What the work involves

You review AI-generated attempts at radiology workflow tasks and judge whether they hold up against how Visage 7 actually behaves. A typical unit of work: a model is given a partially read study handed off mid-worklist and asked what it can conclude, what measurements are preliminary versus part of a signed report, and what follow-up flag belongs on the patient's record. You mark where the output is wrong, explain precisely why in Visage 7 terms — hanging protocol constraints, prior-study linking, MPR versus source-series measurement, addendum versus original signed report — and often write the corrected reference answer yourself. Some tasks ask you to author the scenario first: a realistic case with a defensible single correct handling, plus the distractors a plausible-sounding model would fall for.

A lot of the value sits in the distinctions the listing keeps returning to. Confirmed by this study versus assumed from a prior report. Working annotation versus finalized entry. Graded under the TI-RADS or O-RADS version current as of the exam date, not the version you happen to know best. Traceable in the audit trail versus noted informally. Models are fluent at all of these and reliably wrong on the edges, which is why the screening probes them hard.

What the platform screens for

The screen is AI-led and follow-up heavy. Expect to name your institution's Visage build behavior, describe hanging protocols you or your team configured, and walk a specific case from worklist assignment through signing and addendum. Vague seniority claims collapse fast; concrete recall of where a setting lives and what it constrains does not. You will also be given at least one flawed model answer and asked to critique it — the screen is measuring whether you can articulate the error mechanism, not just flag that something feels off.

Logistics

Fully remote, asynchronous; tasks pulled from a queue rather than scheduled shifts
Commitment typically 10–20 hours/week, with some projects accepting less
Paid hourly at observed rates of $100–130/hr; bands vary by credential, specialty, and project and are not guaranteed
No patient data is used — cases are constructed or de-identified
Engagements are project-based and can ramp up or pause with client demand

Domain Expert – Radiology (Visage 7 Research)

What the work involves

What the platform screens for

Logistics

Interview prep — 10 questions