ML Engineer – Infrastructure

About the role

Prior Labs is building tabular foundation models and spends tens of millions of euros each year on GPU compute. The ML Engineer, Infrastructure will own the multi‑cluster GPU infrastructure, make high‑impact decisions on architecture, scheduling, and hardware, and ensure researchers can train state‑of‑the‑art models efficiently.

Key responsibilities

• Own and evolve multi‑cluster GPU infrastructure, currently Slurm on GCP, expanding to multi‑provider and new hardware.
• Drive GPU utilization and training throughput through profiling, memory optimization, and debugging distributed training.
• Architect next‑generation infrastructure, including multi‑cluster orchestration, new GPU generations, and capacity planning.
• Build developer productivity tools such as CI pipelines, experiment tracking, model registry, and data‑processing tooling.
• Manage the compute budget, understanding cost per FLOP across providers and eliminating waste.

Required profile

• 3+ years building and operating production GPU infrastructure or distributed training systems at scale.
• Deep hands‑on experience with Slurm and cluster management, including debugging scheduling failures.
• Expert‑level systems thinking about memory bandwidth, GPU profiling, and hardware performance.
• Strong Python skills and fluency with PyTorch internals for profiling training runs.
• Proven track record of infrastructure decisions that improved throughput or cost efficiency.
• Experience with AI tooling such as Claude Code, Cursor, or similar.

Required skills

• Slurm
• Google Cloud Platform (GCP)
• Docker
• wandb
• GitHub Actions
• uv
• Python
• PyTorch
• Triton
• CUDA