What the work involves

You'll work on the document-and-deck side of model training for a foundational AI lab. In practice that means three recurring task types: writing reference artifacts from a prompt (a lease abstract summary from a full lease, a banquet event order from a client brief, a run-of-show for a 400-guest gala), grading two or more model attempts at the same artifact against a rubric, and rewriting weak model output into something you'd actually send to a client or an operations team. Files come in as Word, Excel, and PowerPoint — formatting, formula integrity, and slide hierarchy are part of what's being judged, not decoration around it.

The subject matter spans all three verticals, though most contributors anchor in one. Typical assignments include property listing packages with comps and financials, venue sourcing proposals with cost breakdowns, guest experience service standards documents, rent rolls and abstract tables, and event schedules with load-in, vendor, and cue-to-cue detail. Your written justification for a rating usually matters more than the rating itself — the lab is buying your reasoning about why a lease abstract missed an assignment clause or why a BEO with no set count is unusable.

What the screen looks for

  • Specific production history. Not "worked in events" but which documents you personally produced, for whom, at what scale, and in what software.
  • Error-spotting under time pressure. You'll be asked what's wrong with plausible-looking output. Vague dissatisfaction reads as inexperience; naming the omitted clause, the broken escalation, the missing gratuity line reads as fluency.
  • File craft. Excel beyond basic formulas, decks that survive being presented, documents formatted to a house standard.
  • Written clarity. Rationales are read by researchers who don't know your industry.

Logistics

Fully remote and asynchronous — tasks are claimed from a queue rather than scheduled. Commitment is flexible at 5–20 hours per week, with more available for contributors who stay consistent on quality. The rate observed for this listing is $70/hour; pay on these programs varies by task type and is not guaranteed. Screening includes an AI voice interview covering background, a couple of domain scenarios, and availability.