Senior JGPU Engineer

Position Title: GPU Engineer Client: Agency of the United Nations Project: GPU Workload Optimization and Streamlining Duty station: Offsite Period of assignment: September 2025 to December 2025 Possible extension dependent on business need, budget availability and consultant performance. Offsite Rate: Based on Experience Organizational context A specialized agency of the United Nations headquartered in Rome, Italy, with over 40 offices globally, is dedicated to transforming rural economies and food systems to be more inclusive, productive, resilient, and sustainable. the Agency invests in vulnerable groups, including small-scale food producers, women, young people, and other marginalized communities in rural areas. As the Agency continues its mission to reach the last mile, technological innovation remains critical. The ICT Division leads the technology stream with innovations. Within this context, the Agency seeks to further develop its automation capabilities by optimizing cloud-based GPU workloads to support strategic use cases, particularly those leveraging AI and data-driven approaches. Technical Context This Terms of Reference defines the scope, objectives, and deliverables for the GPU Engineer role, focused on building, profiling, and optimizing GPU-accelerated workloads. The project involves the design, implementation, benchmarking, monitoring, and tuning of GPU-based solutions, as well as automation of GPU resource deployment and configuration through modern mechanisms and Infrastructure as Code (IaC) methodologies. Duties and Responsibilities Under the supervision of head of PCE and GPU Lead, the GPU Engineer will be required to provide specialized consultancy services including, but not limited to, the following activities to achieve the deliverables outlined below: ·      Conduct comprehensive performance benchmarking, profiling, and tuning of GPU workloads to provide evidence-based recommendations on suitable GPU sharing techniques (e.g., time-sharing, MIG, vGPU, multi-process service [MPS]). ·      Optimize performance of existing and new applications by leveraging GPU parallelization, identifying bottlenecks, and deploying code and framework-level improvements. ·      Perform a thorough analysis of the deployment methods for GPU-accelerated serving frameworks in the market, with reference implementations and best-practice recommendations for large-scale serving solutions (e.g., NVIDIA Triton Inference Server, TensorRT, ONNX Runtime). ·      Develop repeatable and automated configuration templates for GPU resources. ·      Implement active GPU monitoring, including review and analysis of all relevant metrics (utilization, memory bandwidth, power, temperature, etc.), and establish dashboards and alerts for proactive performance and health management. ·      Integrate GPU resource provisioning and configuration into CI/CD pipelines using Infrastructure as Code (IaC) tools (e.g., Terraform, Helm Charts, etc), and document workflows for seamless deployment and rollback. ·      Document all configurations, testing results, benchmarking analyses, and deployment procedures to ensure transparency and reproducibility. ·      Establish active GPU monitoring protocols, including the identification and evaluation of available metrics, to select the most relevant indicators for ongoing performance management. ·      Support self-service deployment of Large Language Models (LLMs) on GPU resources, enabling application owners with varying technical expertise to access and utilize GPU capabilities seamlessly. ·      Provide knowledge transfer and training sessions to in-house ICT staff on GPU workload management, optimization strategies, and troubleshooting. Technical Requirements & Experience ·      Minimum 2 years of hands-on experience in GPU engineering or cloud-based GPU workload optimization, ideally within enterprise or large-scale environments. ·      NVIDIA Certified (Preferred) ·      Direct experience with GPU services, including resource provisioning, scaling, and optimization. ·      Demonstrable expertise in GPU-accelerated software development (CUDA, OpenCL, TensorRT, PyTorch, TensorFlow, ONNX, etc.). ·      Strong background in performance benchmarking, profiling (Nsight, nvprof, or similar tools), and workload tuning. ·      Experience with Infrastructure as Code (Terraform, HELM Charts, or equivalent) for automated cloud resource management. ·      Proven experience designing and implementing CI/CD pipelines for GPU-enabled applications using tools like GitHub Actions(Preferred) or similar. ·      Working knowledge of Kubernetes and GPU scheduling, including setup of GPU-enabled clusters and deployment of GPU workloads in kubernetes. ·      Familiarity with GPU monitoring and observability, using tools such as Prometheus, Grafana, NVIDIA Data Center GPU Manager (DCGM), or custom scripts. ·      Proven ability to analyze deployment approaches for GPU-accelerated serving frameworks and deliver reference implementations. ·      Experience implementing software quality engineering practices (unit testing, code review, test automation, reproducibility). ·      Strong scripting skills in Python, Bash, or PowerShell for automation and monitoring purposes. ·      Excellent analytical, problem solving, and troubleshooting abilities. ·      Quick learner, adaptable to evolving requirements and emerging GPU/cloud technologies. ·      Positive and collaborative attitude in Agile environments. Required Competencies ·      Problem Solver: Ability to propose and communicate effective solutions in a clear and understandable manner to technical and non-technical stakeholders. ·      Team Worker: Ability to develop and maintain positive relationships with end users and colleagues; consult and collaborate effectively across a multicultural organization. ·      Planner and Organiser: Ability to prioritize and organize work to resolve functional and technical challenges efficiently. ·      Performer: Proven success in implementing high-quality, scalable, and robust GPU/cloud projects. ·      Accountability: Takes ownership of tasks and delivers outputs within prescribed timeframes, cost, and quality standards; operates in compliance with organizational regulations and rules. Languages The project will be conducted in English.