What the work involves

A foundational AI lab is training a model on professional finance deliverables, and the training signal comes from people who have actually owned a close. Day to day you will do three things in rotation: author reference artifacts from a prompt (a month-end journal entry package, an AP aging with commentary, a budget-to-actual variance summary for a department head), evaluate model-produced versions of the same deliverables, and rewrite weak outputs into something you would hand to a controller or an audit senior.

The evaluation half is where judgment shows. A model spreadsheet can foot perfectly and still be wrong — a reconciling item parked in a plug, a tie-out that references a hardcode instead of the trial balance, a PBC schedule that answers a different request than the one the auditor sent. You will be asked to say what is wrong, how materially wrong it is, and what a competent preparer would have done instead, in writing that a reviewer who is not an accountant can follow.

What the platform screens for

  • Hands-on ownership, not oversight. Screeners probe for reconciliations you prepared and schedules you sent to auditors, with specifics: which accounts, what volume, what broke.
  • GAAP and control-testing depth under follow-up. Expect second and third questions on the same topic — accrual cutoff, SOX walkthrough documentation, sampling, segregation-of-duties exceptions.
  • Spreadsheet mechanics. Pivot tables, XLOOKUP/INDEX-MATCH, structured tie-outs, and how you'd audit someone else's workbook for fragility.
  • Calibrated critique. Whether you can separate a formatting nit from a materially misstated schedule and rank them accordingly.

Logistics

Fully remote and asynchronous, with no fixed hours or standing meetings. Commitment is flexible at 5–20 hours per week, more if you want it, and most contributors fit it around a full-time role. The published rate is $80/hour as observed on the platform; rates on AI training work vary by task type and are not guaranteed. Screening includes an AI voice interview covering background, domain depth, and availability.