Deloitte

AI Supercompute Infrastructure Solutions Architect | Poland

Brak informacji o wynagrodzeniu

SeniorHybryda·WarszawaFull-time

#457949·Dodano 21 dni temu·4

Źródło: nofluffjobs.com

Aplikuj teraz

Tech Stack / Keywords

AIGPUStorageNetworkingFabricLinuxCloudKubernetes Node.jsCommunication skillsInfrastructure as CodeAnsible Terraform Python BashGitOpsSwitchesEthernetKubeflow MLflowNetAppLLMRAG

Firma i stanowisko

We are a technology consulting firm building and operating next-generation AI supercompute infrastructure for the world’s most ambitious organizations. We are a repeatedly awarded NVIDIA Consulting Partner of the Year in EMEA, holding one of the deepest and most recognized NVIDIA partnerships in the region, providing privileged access to adoption programmes, reference architectures, and NVIDIA engineering teams.

Wymagania

At least 4-8 years of hands-on experience in infrastructure, systems, platform, networking, cloud, HPC, storage, automation, or data-centre engineering.
Strong practical hands-on depth in at least one core infrastructure discipline such as GPU/server infrastructure, enterprise or high-performance networking, storage, Linux and cluster platforms, cloud infrastructure, automation, or data-centre facilities.
Strong interest in developing breadth across compute, storage, networking, physical data-centre infrastructure, and cluster platforms.
Ability to collaborate with specialists in adjacent domains, ask appropriate questions, understand cross-stack dependencies, challenge assumptions constructively, and produce coherent end-to-end solutions.
Experience designing, deploying, operating, or supporting enterprise-grade server, storage, networking, private-cloud, HPC, or data-centre environments.
Strong understanding of server architecture including CPU, GPU, memory, PCIe, NVMe, NICs, BIOS, firmware configuration, remote management, hardware lifecycle, and fault isolation.
Good understanding of data-centre fundamentals such as rack design, power distribution, redundancy, cooling, structured cabling, physical installation, and operational readiness.
Solid knowledge of networking fundamentals including switching, routing, VLANs, MTU, link aggregation, IP addressing, traffic flows, latency, bandwidth, resiliency, and fault isolation.
Working knowledge of high-performance networking technologies like InfiniBand, RDMA, RoCE, GPUDirect RDMA, or equivalent HPC interconnects.
Understanding of storage concepts relevant to AI/high-performance workloads including block, file, object, parallel file systems, NVMe storage, throughput, IOPS, metadata, replication, backup, and data lifecycle.
Familiarity with Linux administration and troubleshooting including system services, networking, storage, drivers, package management, logs, diagnostics, and automation.
Practical familiarity with Kubernetes, Slurm, or equivalent cluster platforms including node onboarding, GPU scheduling, workload placement, networking, persistent storage, and day-2 operations.
Experience working with technology vendors, delivery partners, customers, or cross-functional engineering teams.
Ability to translate incomplete/ambiguous requirements into technical decisions, architecture principles, design artefacts, implementation plans, and measurable acceptance criteria.
Strong written and verbal communication skills to explain technical trade-offs to engineering teams, vendors, customers, and executives.
Comfort working in data-centre environments reviewing physical infrastructure, server hardware, racks, cabling, power, cooling, and installation quality.
Experience with infrastructure automation or Infrastructure as Code tools such as Ansible, Terraform, Python, Bash, GitOps.
Fluent English and Polish required.

Bonus experience:

Experience with NVIDIA AI infrastructure (DGX, HGX, RTX PRO Servers, BasePOD, SuperPOD, DGX SuperPOD, NVIDIA AI Enterprise, reference architectures).
Exposure to NVIDIA networking products like Mellanox, ConnectX NICs, BlueField DPUs, Quantum InfiniBand switches, Spectrum Ethernet switches, LinkX cabling, NVIDIA UFM, or NVIDIA NEO.
Familiarity with AI platform components such as Kubernetes, Slurm, NVIDIA GPU Operator, DCGM, NVIDIA Base Command Manager, Run:ai, Ray, Kubeflow, MLflow, vLLM, Triton Inference Server.
Experience designing high-performance storage environments including parallel file systems, NVMe-over-Fabrics, Lustre, DDN, VAST Data, NetApp, Pure Storage.
Familiarity with GPU workload characteristics, distributed training, collective communication patterns, NCCL, inference serving, large language model fine-tuning, RAG pipelines, and data-pipeline bottlenecks.
Experience with liquid-cooled infrastructure, direct-to-chip cooling, high-density GPU racks, or data-centre design for AI and HPC workloads.
Hands-on experience with infrastructure validation or performance tools such as DCGM, NCCL tests, iperf3, perftest, ibdiagnet, fio, MLPerf, or vendor-specific diagnostics.
Experience in consulting, technical presales, solution architecture, RFP delivery, or other customer-facing technical roles.

Obowiązki

End-to-End AI Infrastructure architecture:

Translate customer requirements—including AI training, fine-tuning, inference, RAG, simulation, data volumes, performance targets, availability objectives, and future growth—into complete architectures across compute, storage, networking, data-centre infrastructure, and platform software.

Solution definition and technical leadership:

Lead technical discovery, requirements gathering, architecture workshops, solution reviews, and design decisions.
Establish target-state architecture, identify dependencies and risks, define technical standards, and provide a clear path from concept through implementation.

Bill of Materials development:

Develop detailed, technically validated, and commercially viable BOMs covering GPU servers, CPU and management nodes, storage systems, InfiniBand or Ethernet fabrics, switches, NICs, DPUs, optics, cables, racks, PDUs, power infrastructure, cooling equipment, and relevant software, subscriptions, and support services.

Vendor architecture coordination:

Work with server OEMs, GPU vendors, storage providers, networking vendors, system integrators, and specialist partners to validate solution configurations, confirm compatibility, align performance and resiliency assumptions, review lead times, compare options, and resolve technical dependencies.

Compute architecture and sizing:

Define infrastructure for AI training, inference, fine-tuning, LLM Inference, and data-processing workloads.
Size GPU capacity, GPU memory, CPU, RAM, topology, local NVMe, host networking, management infrastructure, and expansion paths.

Storage architecture:

Define high-performance storage architectures appropriate for AI and GPU-intensive workloads covering capacity, throughput, IOPS, metadata performance, caching, tiering, data protection, backup, replication, and integration with Kubernetes, Slurm, and other cluster platforms.

Network Fabric architecture:

Define high-performance InfiniBand and Ethernet/RoCE fabrics for GPU clusters.
Validate topology, design targets, switch placement, host connectivity, optics, cabling, routing, QoS requirements, congestion management, and operational needs.

Data Centre and Facility integration:

Act as technical liaison among customers, data-centre operators, colocation providers, facilities teams, installation contractors, and technology partners.
Define and validate rack layouts, floor-space, power density, cooling, cabling pathways, and commissioning plans.

Physical deployment assurance:

Define installation standards, review rack elevations and port maps, validate cabling, oversee teams and contractors, and ensure physical implementation matches approved architecture.

Platform Integration:

Partner with platform engineering to ensure infrastructure readiness for Kubernetes, Slurm, GPU scheduling, workload orchestration, multi-tenancy, identity and access, observability, security, node lifecycle management, and automation.

Design validation and technical acceptance:

Define acceptance criteria and coordinate validation of hardware health, network fabric, storage performance, node configuration, benchmark execution, burn-in testing, documentation, and handover.

Customer, presales, and delivery support:

Create architecture and design documents, BOMs, rack elevations, network diagrams, deployment plans, technical proposals, and statements of work.
Lead workshops, support RFPs and customer presentations, and provide technical leadership throughout the solution lifecycle.

Deloitte

84 aktywne oferty

Zobacz wszystkie oferty

Aplikuj teraz