Remote ML Ops Engineer — Scalable GPU Inference

Pragmatike; Madrid; A distancia; EUR 90.000 - 120.000; Jornada completa

Descripción de la vacante

Pragmatike is seeking an ML Ops Engineer to join its fully remote, Europe-wide team. You will build and operate scalable ML inference platforms powering real-time AI applications, with a strong emphasis on reliability, low latency, and cost efficiency in GPU-driven environments.

In this hands-on role you will design and maintain production-grade model serving infrastructure (vLLM, TGI, Triton), implement robust deployment pipelines (blue/green, canary), and drive observability, CI/CD, and

Formación

• 4+ years in ML Ops or similar infrastructure roles.
• Production-grade model serving experience.
• Experience with GPU-based inference workloads.
• Strong Python and IaC skills.
• Remote-first with ownership mindset.

Responsabilidades

• Build production-grade model serving infrastructure using vLLM, TGI, Triton.
• Design deployment pipelines with blue/green and canary rollouts.
• Develop auto-scaling and multi-model serving architectures.
• Optimize GPU utilization, memory, network, and storage.
• Design observability for latency, throughput, costs, health.
• Manage model registries and CI/CD for deployments.
• Own full lifecycle of ML systems including on-call support.
• Define engineering best practices for scalable platform.

Conocimientos

Python

Distributed systems

ML Ops

Remote collaboration

On-call support

Herramientas

Terraform

Kubernetes

Helm

vLLM

TGI

Triton

CI/CD pipelines

Descripción del empleo

Pragmatike is seeking an ML Ops Engineer to join its fully remote, Europe-wide team. You will build and operate scalable ML inference platforms powering real-time AI applications, with a strong emphasis on reliability, low latency, and cost efficiency in GPU-driven environments.

In this hands-on role you will design and maintain production-grade model serving infrastructure (vLLM, TGI, Triton), implement robust deployment pipelines (blue/green, canary), and drive observability, CI/CD, and