What the work actually involves

You will spend most of your time in a browser-based task interface rather than on a console. Typical tasks: read a simulated incident intake — a structure fire report with conflicting caller information, a medical call routed to a fire-first-response unit, a wildland interface incident with limited apparatus available — and produce the dispatch decision a competent telecommunicator would make, with the protocol reasoning written out. Other tasks run the other direction: a model has already generated a call-taking script, a unit assignment, or pre-arrival instructions, and you mark where it deviates from accepted practice, where the wording would confuse a caller, and where it invents a resource or capability that does not exist.

The written component is heavier than most dispatch professionals expect. A model cannot learn from "wrong unit assigned"; it learns from "engine-only assignment is insufficient for a reported working fire in a two-story residential structure — local response standards call for a full first alarm assignment, and the AI's response also omitted any callback or pre-arrival instruction to the reporting party." Reviewers are graded on whether their critique names the specific protocol step or operational reason, not just the verdict.

What the screening looks for

The screen is AI-led and conversational, with follow-up questions that drill into specifics. Expect it to test whether your dispatch experience is real and recent: the CAD system you worked in, the protocol set your center used (APCO EMD/EFD, ProQA, Medical Priority Dispatch, or a locally authored card set), call volume, whether you were fire-only, combined fire/EMS, or a consolidated PSAP, and whether you ever trained or QA'd other telecommunicators. It also tests evaluation judgment — whether you can distinguish "different from how my center did it" from "actually unsafe or non-compliant," which is the single most common failure mode among strong operational hires.

Logistics

  • Fully remote, contractor engagement, asynchronous task queues rather than shifts
  • Observed band $30–55/hr, varying by task type and demonstrated review quality; rates are as reported by contributors, not guaranteed
  • Volume fluctuates with project phase — many contributors treat it as part-time alongside existing work
  • Requires a reliable computer, stable connection, and comfort typing structured written feedback at length
  • No AI or machine learning background needed; the domain knowledge is the qualification