We are looking for a Performance ML Engineer to optimise the efficiency and cost of our machine learning models, preparing them for large-scale deployment.

Currently, we operate with 8x H100 GPUs in a single-node configuration, which serves our immediate research and development needs, but we anticipate scaling up to clusters of 64+ H100 in the coming years, with compute needs reaching millions of GPU hours annually.

Your mission: Maximise model performance (latency, throughput, cost per inference) through advanced optimisation techniques (quantisation, distillation, CUDA/Triton kernels, GPU profiling), while preparing our architecture for future scalability.

What You’ll Do

• Designing systems for distributed training and inference:

  • Build scalable distributed training pipelines

  • Debug foundational model trainings

  • Oversee and scale up GPU clusters

• Optimisation:

  • Profile GPU usage and bottlenecks

  • Optimise models, implementing techniques such as quantisation, distillation or kernel improvements

• Collaboration:

  • Work with the R&D team to integrate optimisations into production pipelines.

  • Document benchmarks and performance gains (latency, cost, accuracy).

  • Stay up to date on new architectures (e.g., H200, TPU).

Required Skills

• Model Optimisation Experience: Quantisation, distillation, or kernel optimisation (CUDA/Triton).

  • PyTorch Mastery: Deep understanding of internals (memory, attention, custom ops).

• GPU Profiling: Experience with Nsight Systems/Compute or similar tools (e.g., PyTorch Profiler).

• Distributed training: DDP, FSDP.

• Mixed Precision: FP16/BF16, loss scaling, and troubleshooting.

• Problem-Solving: Ability to diagnose performance issues and propose innovative solutions.

• Fluent English (the team speaks English in the day-to-day)

Bonus Skills

• Experience with Triton or custom CUDA kernels.

• Knowledge of ML compilers (e.g., Apache TVM, TensorRT).

• Experience with large-scale GPU clusters (even modest ones).

• Open-source contributions or publications on model optimisation.

• Experience with one or more of these domains: neurology, multi-modality, (conditional) generation or interpretability.

Beyond Technical Skills:

While technical excellence is critical, we place equal importance on how we work together. We believe the best teams are built on:

• Integrity & Respect

  • We are striving for honesty, kindness, and fairness. We value people who treat others with dignity and foster an environment where everyone feels heard.

• Open Communication & Humility

  • Great ideas come from collaboration. We look for teammates who listen actively, communicate clearly, and approach challenges with self-awareness and humility.

• Psychological Safety & Camaraderie

  • We strive to create a space where people feel safe to take risks, ask questions, and grow.

• Prescreen with Paul (Head of People)

• Technical Screen with one Research Engineer

• On-site (Take-home exercise and restitution OR onsite interviews + Behavioural interview)