MLOps Engineer (GPU Training & Inference Infrastructure)

Job Title

MLOps Engineer (GPU Training & Inference Infrastructure)

Location

Seoul, South Korea, on-site.

Company Description

The client is a Seoul-based medical AI startup founded in 2023 by a team from Seoul National University College of Medicine. It builds AI that reads EEG (brain-wave) data to quantify mental health and predict how patients respond to psychiatric and neurological drugs, so that prescribing rests on measured signals rather than subjective judgment alone. The company has built its own EEG measurement device and is now developing world models, biomedical AI, and LLM/agent systems on top of its clinical data.

Its seed round was led by KAIST Ventures, joined by MYSC, GD Ventures, the Impact Foundation, and the former CEO of a listed Korean medical AI company as an angel. In March 2026 it added a strategic investment from Seoul National University Technology Holdings. Next on the roadmap: supporting pharma clinical R&D and clinician decision-making, and AI agents for psychiatry and neurology clinics in the US.

What You'll Do

• Build and run GPU training and inference infrastructure for world models, biomedical AI, and LLM/agent systems.
• Build the ML pipeline and MLOps environment that connects training, evaluation, model registry, and deployment.
• Build experiment management: tracking and reproducing experiment configs, datasets, checkpoints, metrics, and artifacts.
• Schedule, monitor, and optimize GPU resources across on-premise and cloud.
• Build and harden API serving, batch inference, and online inference for LLM/agent and AI models.
• Build post-training pipelines (SFT, preference learning, RL) and agent evaluation environments.
• Work with AI scientists, data engineers, and software engineers to turn research code into reproducible, scalable systems.

Who You Are

• Hands-on experience in MLOps, ML platform, or backend/infrastructure engineering.
• Python-based ML or backend system development.
• Docker and Kubernetes: you have built and operated them, not just used them.
• You understand training and inference workflows in PyTorch or a similar framework.
• You run Linux and GPU-based ML infrastructure and can debug it when it breaks.
• You can take a researcher's requirement and turn it into a working system.
• Stack: PyTorch, Python, AWS, Docker, CUDA, Kubernetes, Kubeflow, Grafana, Prometheus, Terraform, MLflow, GitHub Actions.

Nice to Have

• Experiment tracking and MLOps platforms: MLflow, Weights & Biases, Kubeflow.
• NVIDIA GPU clusters, multi-GPU or distributed training: PyTorch DDP, DeepSpeed, FSDP, Ray.
• GPU model and LLM serving with vLLM or Triton Inference Server.
• LLM/agent evaluation, SFT, preference learning, or RL post-training pipelines.
• Hybrid cloud and on-premise ML infrastructure; Airflow, Argo Workflows, or Kubeflow Pipelines; Terraform, CI/CD, observability.

Compensation & Benefits

• Competitive salary and stock options.
• Meals, in-office amenities, and a self-development allowance.

Process

1. Coffee chat with the CEO
2. Coffee chat with the team (if needed)
3. Technical interview
4. Reference check

Why Join

• You own the ML infrastructure of a company whose models answer real clinical questions, from the first GPU to production serving.
• A medical team from Seoul National University and investors from KAIST and SNU behind the science.
• Small team, early stage: the systems you build become the company's standard.