The work

You are writing the ground truth for how a model learns what "polished" means in a deck. Day to day that means taking reference decks, brand kits, and raw content inputs and turning them into well-specified design tasks — then producing the gold-standard slide set that answers the task at a craft level you would ship to a board. Alongside each task you write the rubric: what counts as on-brand, what counts as a hierarchy failure, how much grid drift is tolerable before an output fails.

A large share of the job is critique with reasons attached. You'll review model outputs and name the failure modes specific to generated design — optical misalignment that passes a snap-to-guide check, type scales that collapse into two indistinguishable sizes, icon sets mixed across stroke weights, accent colors that are near-miss rather than brand. Vague feedback is not usable training signal, so the standard is concrete, reproducible language a second reviewer could apply and reach the same verdict.

What the screen looks for

  • Portfolio evidence of high-craft deck work — board, pitch, sales, or brand-system decks, not general graphic design samples.
  • Style inference — can you look at a reference deck and reconstruct its type scale, grid, and spacing logic accurately.
  • Tool depth — expert PowerPoint and/or Keynote, including master and template mechanics, plus Figma or Adobe familiarity.
  • Rubric thinking — whether your quality judgments translate into criteria someone else can score against.

No prior AI experience is required; domain fluency is the credential being tested.

Logistics

Remote and asynchronous, with a preference for NAMER or Europe-based contractors. Pay is per accepted task rather than hourly — the $20–60/hr band reflects observed effective rates and will vary with your speed and rework volume. A weekly minimum submission applies. micro1 typically closes these roles within about 48 hours and expects first tasks within 24–48 hours of onboarding, so treat this as a start-now engagement rather than a pipeline application.