Write synthetic multi-turn dialogues in which a user and a tool-using AI assistant work through tasks in apps like calendar, email, maps and drive — playing both sides, including the assistant's function calls and simulated tool responses. Turing runs this for a frontier lab; the work is contractor, fully remote, 20+ hours a week with four hours overlapping PST, on a 12-week contract.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

What the work actually is

You are not chatting with a model. You are authoring a transcript: a user with a plausible goal, an assistant that decides when to call a tool and when to just talk, the JSON-shaped function calls themselves, and the tool returns that come back. The interesting cases are rarely the happy path — a task the assistant cannot complete with the tools it has, an ambiguous request that needs one clarifying question rather than three, a tool that returns an empty result or a stale record, a user who changes their mind on turn four. Each conversation has to be internally consistent, correctly formatted against an internal playbook, and defensible when a reviewer asks why the assistant did what it did on turn three.

Expect volume with revision. Batches come back with feedback on realism, argument construction, or formatting drift, and you iterate. Consistency across your own output matters as much as any single strong example — a hundred dialogues that follow the schema beat twenty brilliant ones that each interpret it differently.

What the screen is looking for

API and schema literacy. You can read a function signature and know what a valid call looks like, what the required arguments are, and what a realistic error or empty response looks like coming back.
Judgment about tool use. Knowing when the assistant should not call a tool is tested more than knowing when it should.
Writing that sounds like people. User turns that read like a real person typing quickly, assistant turns with consistent tone and no filler.
Guideline discipline. Evidence you can hold a format across dozens of items without inventing your own conventions.
3+ years professional experience in a technical or analytical field; any stack is fine.

Logistics

Fully remote contractor engagement, no benefits, stated as 12 weeks. Minimum 4 hours per day and 20 hours per week, with at least 4 hours overlapping PST. Hiring is restricted to India, Pakistan, Nigeria, Egypt, Ghana, Bangladesh, Turkey, Brazil and Mexico. Pay is not disclosed in the listing; Turing typically sets an hourly rate at offer stage and rates vary by project and location — treat any figure you hear secondhand as unverified.

LLM Trainer - Agent Function call

What the work actually is

What the screen is looking for

Logistics

Interview prep — 10 questions