What the work actually involves

Project Iter is a rotating set of research tasks scoped to your field rather than a single fixed assignment. In practice that means one week you may be synthesizing a literature base into a defensible summary with citations that hold up under checking, and the next you may be writing problems that a frontier model should be able to solve — then documenting the correct reasoning path, the plausible wrong turns, and why they are wrong. Other tasks lean toward critique: reviewing a model's output against your field's standards, curating or cleaning a domain dataset, or designing an experiment that distinguishes real capability from pattern-matching. Deliverables are written artifacts. Reproducibility matters more than elegance: someone else on the project needs to follow your method and reach your result.

What the platform screens for

The screen is AI-led and probes depth rather than pedigree. Expect to name your subfield precisely, then defend a specific claim within it — the follow-ups get narrower, and vague answers collapse quickly. Two things carry disproportionate weight: whether you can identify a question in your field where a confident-sounding answer is actually wrong, and whether you can articulate your own boundary of competence. The stated gates are a PhD (completed or in progress) in any quantitative, scientific, technical, or humanities discipline, plus at least five years of applicable industry experience. Prior AI/ML exposure, teaching or TA history, peer review or grant writing, and first-author publication are all treated as advantages rather than requirements.

Logistics

  • Fully remote and asynchronous — no standing meetings, no timezone constraint
  • 10–20 hours per week, on projects you choose to accept
  • Engagement window of roughly 2–3 weeks per project, with the possibility of further rounds
  • Pay observed at $120–180/hour, project-based; the rate for any given project is set at assignment and is not guaranteed across the band
  • Early-stage Y Combinator-backed company, so scoping and instructions can shift mid-project; comfort with ambiguity is genuinely part of the job