Forward Deployed Infrastructure Engineer - APAC

Remote. Posted 8 days ago

The role

You will build, validate, and benchmark multinode GPU clusters for reserved customers. You will act as a technical liaison between customers and providers to debug hardware, fabric, and performance issues.

What you will do

Build, validate, and benchmark new customer clusters and provide documentation.
Debug multinode and NCCL performance, GPU and hardware faults, driver and fabric issues, and container and scheduler problems.
Coordinate provider escalations including maintenance windows, RMAs, and disputed fault attribution.
Provide embedded ownership for named accounts by learning their workloads.
Own monitoring and alerting using Grafana and Prometheus.
Automate repeat work and create runbooks for L1 tier.
Act as a technical liaison to prove where faults live and coordinate fixes with providers.

Required qualifications

Deep Linux experience and total comfort in the CLI.
Hands-on multinode GPU experience including NCCL, InfiniBand or RoCE, collective performance debugging, topology and placement.
Hardware fault triage on GPU nodes including XID and ECC errors, dmesg, lspci, nvidia-smi, thermal and power faults.
Experience provisioning and operating GPU clusters with Kubernetes, Slurm, or both.
Comfortable customer-facing, including delivering bad news.
Sound judgment on when to keep digging and when to escalate.

Nice to have

Experience with parallel filesystems (Weka, Lustre, GPFS) and high-performance storage.
Grafana, Prometheus, or similar observability stacks in production.
Prior forward deployed engineer, solutions architect, or technical account manager experience.
Infrastructure as code (Terraform, Ansible) and CI for cluster provisioning.
Exposure to model training or inference workloads from the practitioner side.

About the company

Hyperbolic makes global AI compute liquid by bringing fragmented, heterogeneous GPU capacity together into one unified platform. They aim to make compute flow as freely as electricity.