What the work involves

You will run personalized prompts against Gemini using your own connected Google account and judge what comes back. A prompt might ask the model to summarize a thread, find a photo from a specific trip, reconcile a calendar conflict, or pull a document you actually own — and your job is to decide whether the retrieval was correct, whether the personalization was warranted, and whether the response would genuinely help the person it was written for. Much of the day is side-by-side comparison: two candidate responses, one verdict, and a written rationale that a model trainer can act on. Korean-language prompts and content are central to this cohort, so you should expect to evaluate Korean output and Korean-language personal data alongside English.

The failure modes you're paid to catch are subtle. A response can be fluent, well-formatted, and confidently wrong about who someone is, invent a relationship between two calendar entries, surface the wrong "Jenny" from your contacts, or quietly assume a preference you never expressed. Unsupported assumptions and over-personalization are as much a defect as a missed retrieval, and honorific or register errors in Korean output count too.

What the platform screens for

  • Verifiable eligibility first: US residence, and a real, actively used Google account with enough history across apps to generate meaningful retrieval prompts. A sparse or freshly created account cannot support this work.
  • Informed consent: you must be genuinely willing to grant app integration permissions and sign standard consent forms and NDAs. Hesitation here is a legitimate no — better to surface it in the screen than mid-project.
  • Evaluation judgment under follow-up: expect to be asked why a response is wrong, not just whether it is, and to hold a position when the screener pushes back.
  • Korean fluency at a native or near-native level, including sensitivity to register and honorifics in generated text.

Logistics

Fully remote, contractor engagement of up to 16 weeks. The commitment is at least 4 hours per day and up to 40 hours per week, with 4 hours of daily overlap with PST — plan your schedule against Pacific time before applying. You need a desktop or laptop and reliable internet; phone-only setups are not workable. Shortlisted candidates receive a Job Interest Form, and selected candidates move to consent and onboarding. Pay is undisclosed on this listing; ask for the rate and any minimum-hours guarantee in writing before onboarding.