What the work involves

AfterQuery is building a research tool that answers complex questions by generating structured tables — rows of entities, columns of attributes, each cell ideally traceable to a source. Your job is the human layer on top of that. You open a generated table, read the question it was meant to answer, and go cell by cell: does this entry belong in the table at all, is the value correct, does the linked source actually say what the cell claims? Broken links, sources that discuss a related-but-different entity, and numbers that drifted between the source and the cell are the routine failure modes you will be flagging.

The bilingual part matters because much of the source material is Chinese-language — mainland regulatory filings, Taiwanese and Hong Kong media, company sites, academic pages — and the model both misreads those sources and skips them entirely. You will be reading Simplified and Traditional text, judging whether a Chinese source is credible and current, and writing your corrections and notes in clear English. The second half of the task is additive: independent web searching to find rows the AI omitted, which often means searching in Chinese for things an English-first pipeline never surfaced.

What the platform screens for

  • Verified fluency, both directions. Screens typically probe reading comprehension of unglossed Chinese source text and the ability to summarise it accurately in English.
  • Source discipline. Whether you actually click through, and whether you can articulate why one source outranks another rather than treating any URL as validation.
  • Calibration. Reviewers who mark everything wrong are as unhelpful as reviewers who mark everything fine. The qualification assessment is largely a calibration test against a gold-standard table.
  • Independence. There is no supervisor watching the queue; unclear cases need a documented judgment, not a stalled task.

Logistics and pay

Fully remote, asynchronous, project-based — tasks are claimed from a queue rather than scheduled, and volume fluctuates with project cycles. Most contributors report several hours a week rather than full-time load, and there is no guarantee of continuous work between batches. Note a discrepancy worth raising in your own screen: the platform's posted copy advertises $50/hour, while the rate observed by contributors on this board has been around $30/hour. Neither figure is guaranteed; confirm the rate for your specific project before starting the qualification assessment. Payment is per hour of logged review work, and the unpaid qualification assessment is short.