What the work looks like

You will spend most of your time reading model output and deciding whether it would survive a partner's review. That means checking whether a drafted provision actually allocates risk the way it claims, whether a cited standard exists, whether a bankruptcy timeline is right, and whether a compliance conclusion holds under the governing statute. Corrections matter more than scores: the useful deliverable is a rewritten passage plus a short explanation of why the original reasoning failed.

The other half is generative. You will author realistic scenarios — a disputed earn-out, a preference-payment analysis, a conflicted board approval, a Reg FD question — along with research memos and hypotheticals that expose where models reason poorly. You may also help write annotation guidelines and rubrics that other attorneys apply, and occasionally join calls with engineers and program leads to explain why a legal distinction that looks cosmetic is not.

What the screen looks for

  • Verifiable credentials. JD from an accredited US law school and an active bar license in at least one US jurisdiction; expect to name the jurisdiction and bar number.
  • Substantive depth. 3–15 years in litigation, corporate transactions, bankruptcy, or compliance, described at the level of specific deal or matter mechanics rather than practice-area labels.
  • Evaluation judgment. Whether you can distinguish a confidently wrong legal conclusion from a defensible one you happen to disagree with, and whether your feedback is specific enough to act on.
  • Writing under constraint. Clear, structured prose that holds up when compressed into a rubric or annotation field.

Judicial clerkships, law-teaching experience, and law journal editorial work are all treated as meaningful signals here — they correlate with the reasoning-transparency the work demands.

Logistics

Remote, US only, contractor engagement. Work is largely asynchronous with periodic sync calls; hours are flexible but projects typically ask for a stated weekly commitment, often part-time alongside an existing practice. No prior AI or machine-learning experience is required or expected. Volume fluctuates with client project cycles, so treat this as variable rather than guaranteed hours, and confirm any conflicts or outside-activity obligations with your firm or bar before starting.