What the work actually involves

You are not counseling patients here — you are judging whether a model's nutrition output would hold up if a real patient acted on it. A typical task queue mixes a few formats: reading a synthetic client case (labs, medications, diagnosis, food preferences, budget) and writing the assessment and intervention a competent RD would deliver; grading two model-generated meal plans against each other and explaining, in writing, which one is defensible and why; flagging outputs that quietly drift into scope violations, unsupported supplement claims, or advice that ignores a drug–nutrient interaction. Some projects also ask for original reference answers — the gold standard the model is trained toward — which means citing MNT guidance, DRIs, or condition-specific standards rather than personal practice preference.

What the platform screens for

micro1 runs an AI-led interview before any project match. It verifies the RD/RDN credential and state licensure status, then probes clinical depth with follow-ups: you say you work in renal nutrition, it asks how you'd adjust potassium targets for a patient on a specific dialysis modality. The screen is also looking for whether you can articulate why an answer is wrong, not just that it is — rater rationales are the training signal, so vague labels like "inaccurate" carry no value. Expect questions about handling ambiguous cases where the evidence is genuinely mixed, and about staying inside RD scope when a prompt invites you outside it.

Logistics

  • Contractor, 1099, remote, US-based; licensure must be active and unrestricted
  • Asynchronous task queues — you pick up work when you have time, with deadlines per batch rather than fixed shifts
  • Most contributors run 10–20 hours a week alongside clinical practice; some projects offer more volume in bursts
  • Written communication with project coordinators, occasional calibration calls
  • Pay is hourly and observed in the $50–100 range; rates vary by project scope, specialty, and calibration performance, and are not guaranteed

No AI or machine-learning background is expected. The learning curve is the rubric, not the technology.