ML Engineer, Infrastructure

1 month ago

Berlin, Germany or New York, NY, USAMid Level

Responsibilities

Own and evolve multi-cluster GPU infrastructure, including architecture, scheduling, reliability, and cost optimization.
Improve GPU utilization and distributed-training throughput through profiling, memory optimization, and systems-level debugging.
Architect multi-cluster orchestration, provider diversification, hardware adoption, and compute capacity planning.
Build developer productivity tooling for CI pipelines, experiment tracking, model registries, data processing, and internal research workflows.
Manage compute costs and evaluate cost per FLOP across providers and hardware.
Collaborate directly with researchers and contribute to model training when appropriate.

At least 3 years of experience building and operating production GPU infrastructure or distributed-training systems at scale.
Deep hands-on experience with Slurm and cluster management, including multi-tenant GPU workload scheduling and utilization optimization.
Expert systems knowledge covering memory bandwidth, GPU profiling, and hardware-level performance analysis.
Strong Python skills and genuine fluency with PyTorch internals.
A track record of improving training throughput or cost efficiency through infrastructure decisions.
Strong familiarity with AI productivity tooling such as Claude Code, Cursor, or similar tools.
Preferred experience includes tens-of-millions-scale GPU spending, multi-cloud or hybrid HPC/cloud infrastructure, Triton, CUDA, custom kernels, multi-cluster orchestration, and experiment tracking, model registry, or ML pipeline tooling.

Teams are based in Berlin, Freiburg, and New York, with exceptional remote arrangements generally involving frequent travel to an office.
The company brings everyone together regularly for offsites.
The company offers equal opportunities and welcomes applicants from diverse backgrounds.

51-200 employees