Чётко описанная ведущая роль сочетает практическую работу с GPU-инфраструктурой для ML и LLM, современный стек и конкретные возможности развития; в консалтинговой модели EPAM содержание работы может зависеть от запросов клиентов.
Актуальные технологии
Инфраструктура для ИИ
Поддержка развития
?Клиентские проекты
Мэтч & Сопровод
Покажет вашу совместимость и напишет письмо
Создать профиль и узнать мэтч
Описание вакансии
Текст:
Hirify AI/Оригинал
TL;DR
Lead DevOps/MLOps Engineer (AI): Building and hardening an on-premises, GPU-backed ML platform for concurrent ML and LLM workloads with an accent on Kubernetes/OpenShift operations, GPU resource management, and platform reliability. Focus on implementing GPU sharing and isolation, scaling H200 infrastructure, monitoring GPU performance, and supporting MLOps services and Ray-based training.
Location: Remote, hybrid, or onsite in Kazakhstan; offices available in Astana, Almaty, and Karaganda.
Company
EPAM is a global AI transformation engineering and integrated consulting company providing custom software, product, and platform engineering for global enterprises and startups.
What you will do
Build and harden an on-premises, GPU-backed ML platform for concurrent ML and LLM workloads.
Deploy and operate the NVIDIA GPU Operator across Kubernetes and OpenShift clusters, including GPU sharing and isolation strategies.
Troubleshoot resource bottlenecks and latency issues, and scale the GPU environment as H200 servers are added.
Maintain MLflow, the Agent Gateway, monitoring pipelines, and GPU performance observability with NVIDIA DCGM exporter, Prometheus, and Grafana.
Collaborate with Deep Learning, Traditional ML, and Data Analytics teams on shared infrastructure, Ray-based training, and the in-house agentic platform.
Ensure platform reliability and security for confidential on-premises data.
Requirements
5+ years of production experience running GPU workloads on Kubernetes or OpenShift, ideally on-premises; total experience stated for the role is more than 6 years.
Practical experience deploying and troubleshooting the NVIDIA GPU Operator, with understanding of its components.
Understanding of GPU scheduling as distinct from CPU scheduling, including memory limits, utilization semantics, and scheduling limitations.
Knowledge of GPU sharing and isolation methods including MIG, time-slicing, and MPS, with their trade-offs.
Solid Linux fundamentals, including cgroups, namespaces, systemd resource controls, and process-level resource management.
Experience with Prometheus, Grafana, and NVIDIA DCGM, including process- or pod-level GPU metrics.
Working knowledge of the CUDA stack, including drivers, CUDA Toolkit, and container-runtime version compatibility.
Experience deploying with Helm and operating MLOps tools such as MLflow and Airflow on Kubernetes.
Nice to have
Familiarity with LLM serving engines such as vLLM, SGLang, or NVIDIA Triton.
Experience with Ray, KubeRay, and GPU-aware schedulers such as Kueue or Volcano.
Hands-on experience with H100/H200-class hardware and GitLab CI.
Culture & Benefits
Remote, hybrid, and office work options in Kazakhstan, with the opportunity to work abroad for up to two months per year.
Relocation opportunities within offices in 55+ countries.
Leadership development, career advising, soft-skills and well-being programs, certifications, internal learning resources, and free English classes.
Employee Stock Purchase Plan, referral bonuses, medical and family care, six trust days per year, psychology-session coverage, and additional benefits.
Housing support through preferential mortgage access via the Otbasy Bank partnership.
Будьте осторожны
Если работодатель просит войти в их систему, используя iCloud/Google, прислать код/пароль, запустить код/ПО, не делайте этого. Это мошенники. Обязательно жмите «Пожаловаться» или пишите в поддержку. Подробнее в гайде →