What the work involves

You are producing and judging code that becomes training and benchmark data for frontier models. On a given day that can mean authoring a task with a precise reference solution, correcting model output that compiles but quietly mishandles concurrency or edge cases, or writing a structured rationale explaining why one of two plausible-looking implementations is worse on scalability grounds. A second strand of the work is more architectural: hypothesising about stages of the engineering cycle — prototyping, architecture and API design, production implementation, launch, experimentation, monitoring, maintenance — and building tasks that test whether a model can actually reason about them. You may also be asked to build verifier agents and automated checks that decide, without a human in the loop, whether a candidate solution to a task is correct.

What the platform screens for

Turing's screen is specific about pedigree: several years of engineering experience with at least two continuous years full-time at a top-tier product company. Expect that to be probed by name, by team, by what you actually shipped. Beyond the résumé, screeners look for multi-language fluency that is real rather than nominal — being asked to switch from a Rust ownership question to a React rendering question in the same conversation is normal — and for the ability to articulate a code review verdict in writing, with reasoning a researcher can audit. Candidates who can only say "this code is bad" without naming the failure mode, the input that triggers it, and the fix tend not to progress.

Logistics

  • Remote contractor engagement, no medical or paid leave.
  • Minimum 10 hrs/week, scalable to 40; partial PST overlap is required, so fully async work is not on offer.
  • Initial term is one month with a near-immediate start, and extensions are decided on performance and fit.
  • Pay is undisclosed on this listing; coding evaluation rates at this seniority are typically negotiated per engagement, so ask directly rather than assuming a published band applies.