What the work involves
You spend most of your time reading model outputs and deciding whether they would survive contact with a real client. That means checking whether a purported Anti-Kickback safe harbor actually applies, whether a HIPAA analysis correctly distinguishes a covered entity from a business associate, whether a Stark exception is cited with the right structural elements, and whether state telehealth licensure rules were flattened into a false national answer. When an output is wrong, you write out why — not a rating, but a diagnosis a non-lawyer reviewer can follow.
The other half is generation. You draft prompts and reference answers that mirror problems you have actually handled: a compliance question from a hospital CFO, a records request that mixes 42 CFR Part 2 with state law, a marketing arrangement that sits close to the line. Task types shift as the customer's needs evolve — some weeks lean on annotation of regulatory documents, others on adversarial prompt writing or head-to-head comparison of two model responses.
What the screen looks for
micro1's process is AI-led and conversational. It probes for a real license and real matter history — expect follow-ups that go a layer deeper than your first answer, asking for the statutory hook, the counterargument, or how the analysis changes under a different fact pattern. Vague fluency reads poorly; naming the specific regulation, the year it changed, and the practical workaround reads well. Evaluators also test judgment: can you tell a confidently wrong answer from a defensible-but-incomplete one, and can you explain the difference in writing.
Logistics
- Fully remote, contractor engagement, W-9 or equivalent
- Largely asynchronous; occasional calibration calls with cross-functional teams
- Volume is variable — many contributors work 5–15 hours a week alongside practice
- Rate depends on subspecialty scarcity and task complexity; the $140–400/hr band reflects observed offers, not a guarantee
- No prior AI or ML experience expected