ML Infrastructure Engineer — Multi-Node GPU Training

Axiōma Search; Greater London; On-site; GBP 108,000 - 180,000; Full time

Job summary

A London-based robotics startup is seeking a Research Engineer to own the scaling of training infrastructure, improving speed and compute efficiency. You will work closely with researchers and tackle distributed systems challenges in a hands-on, early-stage environment.

Strong Python/PyTorch skills and experience with multi-node training are essential, with exposure to NCCL, FSDP or DeepSpeed and AWS cloud GPUs preferred. The role emphasizes ownership and fast learning.

Qualifications

• Strong Python and PyTorch experience with ML infrastructure focus.
• Experience scaling training to multi-node GPU clusters.
• Knowledge of distributed training tech (NCCL, FSDP or DeepSpeed).
• Experience with cloud GPUs, preferably AWS.

Responsibilities

• Build and operate multi-node GPU training infrastructure, including orchestration, scheduling, environments and recovery.
• Improve training reliability to reduce downtime from days to minutes.
• Profile end-to-end and remove throughput bottlenecks.
• Address checkpointing, data loading, storage, networking and image builds.
• Develop observability to track utilisation, performance and cost.
• Collaborate with researchers and allocate compute budget decisions.

Skills

Python

PyTorch

ML Infra

Multi-node

NCCL

FSDP

DeepSpeed

AWS

Debugging

Tools

CUDA

Triton

Job description

A London-based robotics startup is seeking a Research Engineer to own the scaling of training infrastructure, improving speed and compute efficiency. You will work closely with researchers and tackle distributed systems challenges in a hands-on, early-stage environment.

Strong Python/PyTorch skills and experience with multi-node training are essential, with exposure to NCCL, FSDP or DeepSpeed and AWS cloud GPUs preferred. The role emphasizes ownership and fast learning.