About Radian Arc.
Radian Arc provides an infrastructure-as-a-service (IaaS) platform for running cloud gaming, artificial intelligence and machine learning applications inside telecommunication carrier networks. Our teams across the USA, Australia, Central Europe, Malaysia, Singapore and Japan offer telecom operators a GPU-based edge computing platform without the need for capital expenditure, facilitating low latency and improved economics for value-added services and the monetization of 5G investments.
What impact you will have
Mission: Design, build, and operate the AI storage layer powering large-scale GPU infrastructure, enabling datasets, model artifacts, checkpoints, and inference state to be delivered to compute clusters with extremely high throughput and predictable latency.
You will play a key role in architecting and evolving the storage platform across edge and core deployments, supporting the full lifecycle of AI workloads including distributed inference, fine-tuning, and large-scale model training. The role spans multiple storage architectures used across the platform, including hyperconverged storage currently based on StorPool, local NVMe storage for latency-sensitive workloads and edge deployments, and disaggregated AI storage platforms such as VAST Data and Weka.
As the first dedicated storage platform role in the organization, this position combines Staff-level architectural ownership, technical direction, and cross-functional influence with hands-on execution across storage design, deployment, performance engineering, troubleshooting, platform integration, and operational improvement.
A key responsibility of this role is designing and optimizing the storage architecture underlying distributed inference stacks such as NVIDIA Dynamo, llm-d, or similar inference orchestration frameworks. This includes ensuring that storage systems efficiently support inference workloads through optimized dataset access, model artifact distribution, checkpoint handling, and KV-cache persistence. You will design scalable storage systems capable of feeding thousands of GPUs while balancing throughput, latency, resilience, and cost efficiency, and work closely with compute, networking, and platform engineering teams to ensure seamless integration with the platform orchestration layer.
Because this is currently the primary storage platform role in the company, the position is intentionally hybrid: you are expected to operate at L6 / Staff in terms of long-term design, standards, cross-team influence, and platform direction, while also directly executing critical storage work that, in a larger organization, would be distributed across multiple engineers.
What you'll do
- Design scalable AI storage architectures supporting both edge and core deployments.
- Define storage strategies for distributed inference, fine-tuning, and training workloads.
- Architect solutions across multiple storage models:
- Hyperconverged infrastructure such as StorPool,
- Local NVMe storage,
- Disaggregated storage systems such as VAST, Weka, and related architectures.
- Define reference architectures, design principles, and reusable patterns for storage platforms so future deployments follow standards rather than one-off implementations.
- Evaluate trade-offs across throughput, latency, resilience, data locality, cost, and operability, and make clear recommendations to engineering and leadership.
- Influence the long-term storage roadmap, including architecture choices for edge, core, hyperconverged, and disaggregated environments.
- Optimize storage throughput and latency for GPU-heavy clusters.
- Design data locality strategies to minimize dataset movement across the network.
- Benchmark storage performance under real AI workloads.
- Optimize I/O patterns for large dataset ingestion, checkpointing, and model artifact distribution.
- Work directly with compute teams to ensure storage architecture matches the access patterns of distributed training, fine-tuning, and inference frameworks.
- Establish performance baselines and validation methods so storage platforms are tested against realistic AI workload behavior rather than only synthetic benchmarks.
- Implement and maintain CSI drivers.
- Integrate storage platforms with Kubernetes and orchestration systems.
- Integrate block, object, and shared file storage into the platform.
- Design multi-tenant storage architectures supporting isolated workloads.
- Ensure storage capabilities are correctly exposed into platform services, workload orchestration, and lifecycle automation.
- Define standards for how storage should be integrated into Kubernetes-based and platform-managed environments across different deployment models.
- Contribute to the design of exabyte-scale storage platforms.
- Support S3-compatible object storage, distributed file systems, and block storage.
- Integrate storage clusters into heterogeneous customer environments.
- Design storage systems with clear fault domains, lifecycle management approaches, scaling paths, and operational boundaries.
- Define reusable operating patterns for multi-cluster and multi-site storage environments.
- Design the storage architecture supporting distributed inference platforms such as NVIDIA Dynamo, llm-d, or similar frameworks.
- Optimize storage performance for large-scale LLM inference workloads.
- Design efficient strategies for KV-cache persistence and retrieval using distributed storage platforms such as VAST or Weka.
- Optimize storage access patterns for token generation pipelines and high-concurrency inference workloads.
- Ensure inference infrastructure scales efficiently across thousands of GPUs.
- Partner with platform and inference teams to ensure storage design supports evolving inference architectures and avoids becoming a bottleneck in throughput, latency, or concurrency.
- Design high-performance data paths between GPU clusters and distributed storage.
- Optimize performance using technologies such as:
- GPU Direct Storage,
- RDMA / RoCE,
- NVMe-oF.
- Ensure predictable latency for inference serving workloads.
- Define architectural approaches for storage-to-GPU data movement that balance performance gains with operational complexity and deployment practicality.
- Work with technologies such as the following, to maximize data throughput to GPU clusters.:
- RDMA,
- RoCE,
- GPU Direct Storage,
- SPDK,
- NVMe-oF,
- Lead storage performance investigations across hardware, network, OS, filesystem, and workload interaction points.
- Drive systematic tuning of storage paths for large-scale GPU environments and define repeatable validation and benchmarking approaches for future deployments.
- Improve the reliability, durability, and observability of the storage stack.
- Collaborate with operations teams to monitor storage systems using telemetry and metrics.
- Optimize performance, latency, and resilience of storage infrastructure.
- Lead incident response and root-cause analysis for major storage events and chronic performance issues.
- Translate operational pain points and incidents into durable design changes, standards, runbooks, and architectural improvements.
- Establish measurable benchmarks for storage reliability, performance consistency, recovery behavior, and operability across deployments.
- Lead end-to-end engineering delivery of storage infrastructure from architecture and validation through production rollout.
- Support practical implementation of storage platforms in both new deployments and existing environments.
- Validate storage BOMs and architecture assumptions together with infrastructure, compute, and deployment teams.
- Contribute detailed input into datacenter layouts, node profiles, and storage topology decisions.
- Drive scaling strategies, capacity planning, and storage lifecycle decisions.
- Ensure storage changes are executed safely with minimal customer impact.
Act as both the architectural owner and the practical execution lead for critical storage initiatives during the build-out phase of the storage function.
-
- Work closely with compute, networking, platform, DevOps, and operations teams.
- Ensure storage integrates seamlessly into the AI platform architecture.
- Act as the primary storage design authority across the organization, guiding adjacent teams on how storage constraints and capabilities should shape platform decisions.
- Communicate architectural decisions, trade-offs, risks, and operational implications clearly to stakeholders.
- Share knowledge and mentor engineers on high-performance storage design.
- Raise the technical bar by helping adjacent teams better understand storage behavior in distributed AI environments.
- CSI.
- NVMe / NVMe-oF.
- Distributed file systems.
- Object storage.
- Linux storage stack.
- RDMA / RoCE.
- GPU Direct Storage.
- SPDK.
- StorPool.
- VAST Data.
- Weka.
- MinioFS.
- Rook Ceph.
What you'll need
- Strong hands-on experience designing and operating distributed storage systems for high-performance compute environments.
- Proven experience designing storage architectures for large-scale AI inference or training platforms, including dataset distribution, checkpointing, and KV-cache storage patterns.
- Deep knowledge of the Linux storage and I/O stack.
- Strong understanding of AI workload data access patterns.
- Experience optimizing storage for GPU-accelerated workloads.
- Familiarity with Kubernetes storage integrations such as CSI.
- Experience operating large-scale storage clusters.
- Experience owning both architecture and direct implementation in lean or fast-scaling environments is strongly preferred.
The candidate should have deep expertise in designing and operating storage platforms optimized for GPU-heavy environments and distributed AI workloads.
This includes a strong understanding of how training, fine-tuning, and inference systems interact with storage, and how storage architecture affects throughput, latency, concurrency, checkpoint recovery, dataset distribution, and serving performance.
Relevant expertise includes:
- Strong understanding of storage access patterns for distributed inference and training.
- Experience designing storage platforms that support large dataset ingestion and model artifact distribution at scale.
- Practical experience tuning storage architectures for checkpointing, distributed file access, object access, and high-concurrency inference.
- Familiarity with storage patterns for KV-cache persistence and retrieval.
- Experience optimizing data locality and reducing unnecessary network movement between storage and compute.
- Understanding of how storage performance affects large-scale AI frameworks, model-serving systems, and inference orchestration layers.
- Ability to debug complex cross-layer issues spanning:
- Storage hardware,
- Networking,
- Linux kernel and I/O paths,
- Filesystems,
- Object and block storage layers,
- Kubernetes integrations,
- Distributed workload behavior.
- Strong knowledge of storage hardware, NVMe devices, storage fabrics, and high-performance data paths.
- Experience designing storage observability systems.
- Strong ability to act as the senior escalation point for ambiguous, high-impact, and multi-domain technical issues.
- Strong automation skills using Python and/or Bash.
- Experience applying software engineering practices to storage automation and operational tooling.
- Experience building reusable tooling, standards, validation patterns, or lifecycle automation that increase leverage across teams.
- Proven ability to lead complex technical initiatives across teams.
- Comfortable collaborating across engineering, operations, deployment teams, vendors, and platform stakeholders.
- Strong systems-level thinking balancing performance, reliability, scalability, operability, and cost efficiency.
- Demonstrated ability to set architectural direction and drive adoption of engineering standards across an organization.
- Proven ability to lead through technical influence across multiple teams and domains, without relying on formal people management authority.
- Strong mentoring capability and ability to raise the technical level of adjacent engineering teams.
- Able to balance short-term execution needs with long-term platform design, operational sustainability, and cost efficiency.
What we offer
- Attractive compensation package reflecting your expertise and experience.
- A great work environment characterised by friendliness, international diversity, flexibility, and a hybrid-friendly approach.
- You'll be part of a fast-growing scale-up with a mission to make a positive impact, offering an exciting career evolution.
Our job titles may span more than one job level. The actual base pay is dependent on a number of factors, such as transferable skills, work experience, business needs and market demands.
Our inclusive responsibility
Radian Arc is committed to creating a diverse and inclusive environment and is proud to be an equal opportunity employer. All qualified applicants will receive consideration for employment without regard to race, color, religion, gender, gender identity or expression, sexual orientation, national origin, genetics, disability, age, veteran status, or any other protected category under applicable law.
