What the work involves

You review model responses written in Brazilian Portuguese and decide whether they read like something a competent native speaker would actually write. That means catching more than grammar errors: European Portuguese constructions leaking into Brazilian text, over-literal translations from English, mismatched formality (o senhor vs. você vs. tu), regionalisms used where neutral Brazilian would be expected, and idioms that are technically correct but nobody says. Typical task types include side-by-side preference ranking, rubric-scored quality grading, rewriting weak responses into a gold-standard version, and writing original prompts designed to expose model weaknesses in Portuguese.

Much of the difficulty is judgment, not language. You will see responses that are fluent and confidently wrong, and responses that are factually fine but tonally off for the requested context — a formal complaint letter, a WhatsApp message to a friend, a legal summary. You are expected to separate those failure modes cleanly in your written justifications, because the annotation rationale is often as valuable to the model trainer as the score itself.

What the platform screens for

  • Native or near-native Brazilian Portuguese, with evidence you've written extensively for Brazilian audiences (education, journalism, translation, localization, law, medicine, technical writing).
  • Working English comprehension, since guidelines, rubrics, and tooling are typically in English.
  • Ability to hold a consistent standard across dozens of similar items and explain a borderline call in two or three sentences.
  • Any domain specialization on top of language ability — finance, healthcare, engineering, legal — which is where the upper end of the observed band tends to sit.

micro1's screening is AI-led: expect an async interview with follow-up questions that push on specifics, plus a short practical exercise on real annotation-style content.

Logistics

Fully remote and largely asynchronous, with work allocated in batches against weekly deadlines rather than fixed shifts. Contributors commonly commit 10–25 hours per week, though volume fluctuates with project cycles and there is no guarantee of continuous work. Pay is hourly and per-project; the $30–65/hr band reflects observed rates across language-only and specialist tasks, not a promise.