Research Engineer - Training Optimization & Infrastructure

Research Engineer – Training Optimisation and Infrastructure

Location : Berlin - Remote within Europe (±2 hours CET)

Level : Mid to Staff

Package : Competitive salary plus equity

The Opportunity

A Series A generative AI company is hiring a Research Engineer to drive optimisation across training strategy and ML infrastructure. The business builds state-of-the-art audio and music generation models and is backed by a leading generative AI fund. The team includes researchers and engineers from Google Brain, Meta FAIR, Amazon, ETH Zürich, and Max Planck.

Role Summary

You will focus on optimising end-to-end training pipelines for large generative models. This includes GPU-level performance tuning, distributed systems work, and driving efficiency across data, storage, orchestration, and experimentation systems.

Key Responsibilities

• Develop and refine training strategies including parallelism approaches and precision choices for varied model scales and compute profiles
• Profile, debug, and optimise single and multi-GPU workloads using tools such as Nsight
• Improve training pipelines covering data storage, data loading, distributed training, checkpointing, and logging
• Build scalable systems for experiment tracking, model and data versioning, and experiment insights
• Design, deploy, and maintain large-scale training clusters using SLURM

Ideal Experience

• Strong hands-on experience optimising training and inference workloads
• Deep understanding of GPU memory hierarchy and hardware performance limits
• Experience tuning both memory-bound and compute-bound operations
• Knowledge of efficient attention algorithms and their performance implications at different scales

Nice-to-Have

• Experience writing custom GPU kernels and integrating them into PyTorch
• Familiarity with diffusion or autoregressive models
• Understanding of high-performance storage solutions such as VAST
• Experience running SLURM clusters at scale

Why Apply

• Work on frontier audio and music generation models
• Influence training strategy and infrastructure at scale
• Join a high-calibre research and engineering team

#J-18808-Ljbffr