What the work actually involves

Read this listing carefully before you apply: the title says annotator, the body describes infrastructure engineering. Turing routinely staffs frontier-lab projects where the same contractor pool both builds the sandbox environments agentic coding models run in and judges the trajectories those models produce. The stated day-to-day is unambiguous — provisioning GPU node pools on GKE and Compute Engine, standing up and tuning vLLM or Triton Inference Server for production inference, containerised deployment with Docker and IaC, data pipelines on Dataflow or Airflow/Cloud Composer, and the observability, autoscaling and cost work that keeps a GPU fleet from quietly burning budget.

If agentic coding annotation is folded in, it looks like this in practice: running a model through multi-step engineering tasks in a real repo, watching where it misreads a stack trace, invents an API, or declares success on a test it never ran, and writing up the failure precisely enough that a lab can act on it. That judgment is easier to demonstrate if you have actually operated the systems the model is pretending to operate.

What the screen looks for

  • Verifiable production experience, not tutorial familiarity — which GCP services you ran, at what scale, and what broke.
  • GPU-specific operational depth: CUDA and driver versioning, quota requests, node pool taints and autoscaling behaviour on preemptible/spot GPUs.
  • Serving-stack specifics: continuous batching, KV cache and memory pressure, tensor parallelism, throughput versus latency tradeoffs under real traffic.
  • Strong Python used for infrastructure and tooling, not only scripts.
  • Clear written reasoning, since annotation output is read by researchers who never speak to you.

Turing's stated process is a profile review followed by an interview with the delivery team if required; AI-led or structured screens on this platform tend to push hard on follow-ups, so shallow answers surface fast.

Logistics

Remote and global, but not async — the posting requires 8 hours per day with a 4-hour overlap with Pacific time, which is a real constraint if you are in Europe, Africa or South Asia. Contractor status with no medical or paid leave, a two-month contract with an immediate expected start, and extension typically depends on the client project continuing. Pay is undisclosed in this listing; Turing's infrastructure contractor rates vary widely by geography and project, so treat any figure you hear secondhand as unconfirmed and ask the delivery team directly.