AI Infrastructure Engineer

Role & responsibilities :

AI Infrastructure Engineer

Owns the GPU fleet, model serving, the Kubernetes clusters in the US and India, the deploy path, reliability, and cloud cost.

Must have

• GPU-backed model inference run in production on Kubernetes, with on-call ownership of that service.
• vLLM in production: tuned under real load, quantized models (AWQ or similar), concurrency and context-length limits, and upgrades rolled out without an outage.
• CUDA and GPU fundamentals: the memory model, profiling with Nsight or equivalent, reading a utilization graph or an out-of-memory error and knowing what happened, and managing the driver, CUDA toolkit, PyTorch, and vLLM version matrix.
• Kubernetes at depth: GPU node pools, scheduling, device plugins, autoscaling (KEDA or equivalent), and persistent volumes for model caches.
• Ownership of the CI/CD path from build to cluster: container builds, registries, and immutable image tags.
• Azure in production, ideally AKS with GPU node pools, including quota, SKU availability by region, and VNet networking. Deep AWS or GCP experience is acceptable.
• Observability:
  Prometheus and Grafana, alert design, and a record of turning incidents into alerts.
• PostgreSQL operations: pooling, migrations under load, replication or failover.
• Strong Python and shell. Terraform or equivalent infrastructure as code.
• Fluent written and spoken English. Works autonomously with clean, reviewable pull requests.