What impact you will have

Mission: Design, build, and operate the compute orchestration layer powering a GPU-native cloud platform for AI and high-performance workloads. (CloudStack, Kubernetes, Slurm, Argo).

The platform orchestrates GPU clusters supporting large-scale AI training and inference workloads across distributed compute infrastructure. This role bridges the current production platform, based on CloudStack, with the next-generation orchestration architecture built around Kubernetes, modern batch scheduling frameworks, and workflow orchestration systems.

You will be responsible for maintaining and evolving the existing CloudStack-based deployments while actively contributing to the design and implementation of the next-generation compute platform supporting distributed AI workloads. The role combines deep hands-on engineering with ownership of critical orchestration components, including Kubernetes-based compute orchestration, Slurm-based distributed training and batch scheduling, and workflow automation through Argo.

Working closely with networking, storage, and platform engineers, you will help implement the platform primitives that expose GPU infrastructure as a scalable, multi-tenant compute platform.

What you’ll do

CloudStack Platform Maintenance

• Maintain the existing CloudStack code base used in current production deployments.

• Integrate new upstream CloudStack releases into the internal platform fork.

• Perform upgrades of existing customer environments to newer CloudStack versions.

• Design and execute safe upgrade paths for running production environments.

• Troubleshoot orchestration and provisioning issues in existing deployments.

CloudStack Networking & VPC Infrastructure

• Maintain and troubleshoot CloudStack VPC networking

• Work with and understand CloudStack Debian VPC routers

• Manage networking implementations based on:

• Open vSwitch (OVS)

• OVN

• Improve reliability of network orchestration components

• Manage hypervisor implementations based on:

• KVM

• QEMU

• Maintain and evolve the code responsible for QEMU GPU passthrough, including PCI mapping and exposure of L40S, RTX 6000 Pro, and H200 GPUs to virtual machines.

Next-Generation Compute Orchestration

• Design orchestration and scheduling primitives for the next-generation platform based on:

• Kubernetes

• Slurm

• Argo Workflows

• Build orchestration workflows that expose GPU and CPU compute resources to platform users.

• Integrate compute orchestration with storage and networking services.

• Work closely with networking, storage engineers, and platform software engineers to integrate platform primitives.

Kubernetes GPU Scheduling & Cluster Orchestration

• Design and implement Kubernetes-based GPU/CPU scheduling infrastructure for multi-tenant AI workloads.

• Configure and maintain GPU device plugins and resource allocation mechanisms.

• Implement GPU scheduling strategies including:

• GPU partitioning, such as MIG where supported

• Multi-GPU job placement

• Topology-aware scheduling for distributed training and inference.

• Design node lifecycle automation for GPU clusters including:

• Node provisioning

• Node draining

• Workload migration

• Implement Kubernetes scheduling extensions where necessary such as custom schedulers or batch schedulers.

Slurm Integration and HPC Scheduling

• Design and operate Slurm-based HPC scheduling environments integrated with Kubernetes clusters

• Implement Slurm compute partitions mapped to Kubernetes-managed GPU/CPU nodes

• Develop mechanisms to submit distributed training, fine tuning, or batch workloads from platform APIs into Slurm clusters

• Implement support for:

• Multi-node distributed GPU training

• Gang scheduling

• GPU topology-aware scheduling

• Build automation for:

• Dynamic Slurm node registration

• Elastic compute capacity

• Node health monitoring and recovery

• Integrate Slurm job lifecycle events with platform orchestration services

Argo Workflow Orchestration

• Design and implement workflow orchestration using Argo Workflows

• Develop reusable workflow templates for common platform workloads including:

• AI training pipelines

• Data preprocessing pipelines

• Batch inference workloads

• Platform operational workflows

• Implement DAG-based execution pipelines coordinating compute workloads across Kubernetes and Slurm clusters

• Build workflow primitives that expose platform capabilities to users such as:

• Distributed training workflows

• Model evaluation pipelines

• Batch GPU compute workflows

• Integrate workflow execution with platform APIs and platform user interfaces

Distributed AI Workload Orchestration

• Implement orchestration support for distributed AI workloads including:

• Multi-node training

• Distributed inference

• Large model fine-tuning workloads

• Support execution environments such as:

• PyTorch distributed training

• MPI-based workloads

• Containerized training jobs

• Implement mechanisms to coordinate GPU workloads across nodes with low-latency networking

Platform Multi-Tenancy & Resource Isolation

• Design and maintain mechanisms for multi-tenant GPU resource allocation

• Implement quota and fairness policies for compute workloads

• Develop resource isolation strategies across tenants including:

• Namespace isolation

• Compute quotas

• GPU allocation limits

• Integrate compute orchestration with platform billing and metering systems

Technical Stack

Programming languages

• Java, Python + Bash, SQL for CloudStack-related work

• Go for Kubernetes-related components

• Python

Orchestration

• CloudStack

• Kubernetes

• KubeVirt

• Slurm/SUNK

• Argo Workflows

• Kubernetes CRDs and controllers

• Batch scheduling frameworks

Networking

• OVS

• OVN

• Linux networking

• VPC networking

• BlueField networking

Infrastructure

• GPU infrastructure

• Distributed compute clusters

• High-performance networking for distributed AI workloads

What you’ll need

Platform & Distributed Systems

• Proven experience working with large-scale distributed compute environments at a neo-cloud, hyperscaler, or HPC provider.

• Strong experience with CloudStack internals, including extending and maintaining platform functionality.

• Experience operating cloud orchestration platforms in production environments.

• Experience running GPU-heavy infrastructure for AI training, inference, or HPC workloads.

Software Engineering

• Experience maintaining or extending large Java codebases, ideally within infrastructure platforms.

• Strong programming skills in Go and Python, with experience building cloud-native platform components.

• Experience designing and maintaining control-plane services for infrastructure platforms.

Compute Orchestration

• Deep practical knowledge of Kubernetes internals and Slurm scheduling systems.

• Experience building or operating compute orchestration layers for large-scale clusters.

• Familiarity with workflow orchestration systems such as Argo Workflows.

Networking & Infrastructure

• Familiar with virtual networking and distributed networking technologies such as OVS, OVN, VPC networking, RDMA, RoCE, ECMP, EVPN/VXLAN, and leaf-spine fabrics.

• Understanding of GPU virtualization and passthrough mechanisms such as QEMU PCI passthrough and NVIDIA MIG.

• Experience working with GPU infrastructure, including passthrough, NVIDIA MIG, scheduling, and lifecycle management of GPUs in distributed clusters.

Leadership & Architecture

• Able to independently own major compute-orchestration initiatives from design through rollout and operational stabilization.

• Comfortable solving difficult implementation and operational problems across CloudStack, Kubernetes, Slurm, and workflow orchestration; improving orchestration quality through code, automation, and practical design decisions; collaborating effectively across compute, networking, storage, and platform teams; and influencing engineering practices through expertise and delivery.

• Comfortable mentoring peers and improving implementation quality, documentation, operational workflows, and platform reliability within the compute orchestration domain.