AI Enterprise Architect

World Wide Technology is looking for a deeply technical Enterprise Architect who will own the delivery of AI projects end to end from the silicon and data center design that underpins AI workloads, through the software and MLOps stack, to the governance frameworks that make AI trustworthy and defensible at scale.

This is a technical hardware-and-software architect role, not a strategy-only position. The successful candidate operates comfortably across GPU infrastructure, high-performance networking, model training and inference pipelines, and the AI risk/governance disciplines increasingly demanded by regulators and enterprise boards.

The Enterprise Architect will lead technical delivery teams for client engagements, acting as the single point of technical accountability from design through to go-live, while mentoring delivery teams and shaping WWT’s broader AI point of view.

Responsibilities

• Own end-to-end technical delivery of AI/ML engagements: architecture definition, design authority, build oversight, and go-live validation.
• Host and chair Architecture Review Board (ARB) and Technical Design Authority (TDA) sessions for AI engagements, owning governance gates, decision records, and design sign-off.
• Architect AI infrastructure spanning GPU/accelerator compute, high-performance interconnects, parallel/high-throughput storage, and orchestration.
• Design the AI software stack: training and fine-tuning pipelines, distributed training frameworks, inference/serving platforms, MLOps/LLMOps tooling, vector databases, and retrieval-augmented generation (RAG) and agentic architectures.
• Define and AI governance frameworks covering model risk management, responsible AI, data lineage, bias/fairness testing, and regulatory alignment (EU AI Act, NIST AI RMF, ISO/IEC 42001).
• Act as trusted technical advisor to client CTOs, CIOs and Heads of Data/AI on platform strategy, build‑vs‑buy decisions, and AI operating model design.
• Lead technical workshops, architecture design sessions, and proof-of-concept builds with cross‑functional engineering, data science, and security teams.
• Mentor other architects and engineers on AI systems design, uplifting AI depth across the practice.
• Partner with sales and pre‑sales to scope AI solutions, size infrastructure, and validate technical feasibility of proposed architectures.
• Architect integration points connecting AI platforms to existing enterprise networks, third‑party systems, and external or service‑provider–hosted environments (e.g. colocation, managed GPU‑as‑a‑service, external inference endpoints).

Qualifications

• 10+ years in enterprise architecture, infrastructure engineering, or platform engineering roles.
• 5+ years focused specifically on AI/ML systems design and delivery, including at least 2 years working with generative AI/LLM workloads.
• Demonstrated track record leading technical delivery (not just advisory) on enterprise‑scale AI or HPC infrastructure programmes.

Required Skills

• GPU/accelerator architectures: NVIDIA / AMD, including multi‑node scale‑out design.
• Accelerator interconnects: NVLink, NVSwitch.
• High‑performance networking: InfiniBand and RoCEv2 fabric design, 400G/800G Ethernet, rail‑optimized topologies for AI clusters.
• Data center facilities: power density, liquid cooling, and rack‑level design considerations specific to AI compute.
• Parallel and high‑throughput file systems (e.g. Everpure, WEKA, VAST, NetApp) sized for training and checkpointing workloads.
• ML frameworks: PyTorch and TensorFlow at a working, hands‑on level.
• Distributed training: Horovod, DeepSpeed, Megatron‑LM, or equivalent multi‑node training frameworks.
• Inference & serving: NVIDIA Triton, vLLM, TensorRT‑LLM, or equivalent high‑throughput serving platforms.
• MLOps/LLMOps: Kubeflow, MLflow, and at least one hyperscaler ML platform (SageMaker, Azure ML, or Vertex AI).
• LLM fine‑tuning (LoRA/QLoRA), RAG architecture design, vector databases (Pinecone, Milvus, Weaviate), and agentic frameworks (LangChain, LangGraph, Semantic Kernel).
• Data lake/lakehouse architectures, ETL/ELT, and data quality/lineage tooling that feed AI systems.
• Infrastructure‑as‑code: Terraform and Ansible for repeatable, automated provisioning of GPU clusters and AI platform environments; GitOps (ArgoCD) for continuous, declarative platform delivery.
• Kubeflow Pipelines, Apache Airflow, or Argo Workflows to orchestrate multi‑stage training, fine‑tuning, and inference pipelines.
• Slurm, Run:ai, and NVIDIA Base Command Manager for GPU job scheduling; Kubernetes-native GPU scheduling including device plugins and MIG partitioning for multi‑tenant clusters.
• Automated model testing, validation gates, and promotion pipelines (continuous training/continuous delivery) that move models safely from experimentation to production.
• NVIDIA DCGM, Prometheus/Grafana, and related telemetry stacks for GPU utilization, thermal, and cluster health monitoring.
• Production model performance monitoring, data/concept drift detection, and LLM‑specific observability (token usage, latency, cost, hallucination/quality metrics) using tools such as Arize, WhyLabs, or Langfuse.
• Centralized logging (ELK/OpenSearch) and distributed tracing (OpenTelemetry) across data, training, and inference pipelines for end‑to‑end root‑cause analysis.
• API‑based and event‑driven integration of AI platforms with enterprise systems, using REST/gRPC APIs and message/event streaming platforms (e.g. Kafka).
• Designing connectivity between AI/GPU infrastructure and existing campus, data center, and WAN environments, including capacity and latency planning for east‑west training traffic and north‑south inference traffic.
• Integrating on‑premises AI platforms with cloud AI services via dedicated interconnects (Direct Connect, ExpressRoute) and multi‑cloud/hybrid connectivity patterns for distributed training or burst inference.
• Experience architecting connections into external or service‑provider‑hosted environments — colocation interconnects, managed GPU‑as‑a‑service offerings, and third‑party/external inference endpoints.

#J-18808-Ljbffr…