What the work actually is

This is an operations-heavy backend role, not a greenfield build. The pipeline it supports produces training and evaluation data for frontier model work, which means the failure mode that matters is silent: runs that complete but package wrong, QC checks that pass data they shouldn't, pods that stall at 300 concurrent jobs and take a trainer's shift with them. Day to day you'll be monitoring deployments across GCP services, debugging failures across large pod fleets and hundreds of concurrent runs, maintaining the Cloud Run-based runner and final-packaging infrastructure, and keeping the Perplexity-customized QC service healthy and modified as requirements shift.

A meaningful slice of the job is human. Trainers — the people generating the data — will come to you directly with technical questions and operational requests, and you are expected to answer without escalating every time. The listing also asks for written runbooks and documentation of recurring issues, which in a three-month contract means the deliverable is partly the handover.

What the screen looks for

  • Production GCP depth, specifically Cloud Run. Generic cloud experience won't survive follow-up. Expect probes on concurrency settings, cold starts, request timeouts, min-instances, and how Cloud Run behaves differently from GKE for long-running jobs.
  • Python you have debugged in production, not Python you have written. Stack traces from live services, memory issues, async failure modes.
  • Kubernetes, SQL, and distributed-systems reasoning applied to a concrete incident you personally resolved.
  • Evidence of independent ownership — you were the person paged, and you decided what to roll back.

Logistics

Remote, contractor, three months, immediate start, no medical or paid leave. Eight hours per day with a 6–8 hour overlap with Pacific time — Turing states a preference for candidates in US or Latin American time zones, which is really a statement about that overlap rather than about geography. Pay is undisclosed on this listing; Turing typically negotiates an hourly rate at offer stage, so come with a number and a basis for it. Evaluation is a profile review, then an interview with the delivery team if the profile clears.