What the work involves

You build the test cases frontier labs use to find out whether a model can interpret in a legal setting without quietly changing what was said. A typical task starts with a scenario you author — an arraignment colloquy, a deposition objection, an attorney-client explanation of a plea offer, a witness using regional slang under cross — rendered as realistic source utterances rather than textbook sentences. You then write the reference interpretation and, crucially, the reasoning: why cargos and not acusaciones here, why the register stays formal when the speaker's does not, why a false start gets preserved rather than smoothed. The grading half of the job is reading model output and deciding whether it is accurate, whether it holds register, and whether a fluent-sounding rendering has shifted burden, tense, modality, or degree of certainty in a way that would matter on the record.

What the platform screens for

AfterQuery's screen is AI-led and follow-up heavy. Expect it to ask for your certification body, credential type, and the settings you have actually worked in, then push on specifics: a term you have seen mishandled, how you handled a witness's ambiguity on the stand, what you do when the source is genuinely incoherent. Vague seniority claims do not survive the second question. The other thing it measures is evaluation judgment — whether you can articulate why an output is wrong in terms a non-interpreter annotator could apply consistently, rather than just marking it bad. Written English matters here as a working tool, not a formality: your rationales are the deliverable.

Logistics

  • 100% remote, fully asynchronous — no live sessions, no assignment calendar
  • Minimum 10 hours per week; you set the schedule
  • Paid weekly via Stripe; observed band is $60–100/hr, typically varying with language pair, credential level, and task type
  • Ongoing rather than one-off, with volume that fluctuates by language pair

Spanish is the highest-volume pair, but Mandarin, Vietnamese, Korean, Arabic, Haitian Creole, Russian, and ASL work appears as labs broaden coverage. If your pair is rare, credential documentation matters more, not less.