AI Research Engineer - AI Safety Platform

Harnham; Remote; GBP 170,000 - 200,000; Full time

Job summary

A Europe-based deep learning company is seeking an AI Engineer to build and optimise frontier-level LLM models from scratch. This hands-on role involves training large models, optimising distributed GPU systems, and contributing to cutting-edge architectures. The position is 100% remote and offers a competitive salary up to £200k plus equity. Ideal candidates will have experience with GPU training, Triton/CUDA, and multimodal training systems.

Qualifications

• Experience in building and optimising training frameworks for LLMs and VLMs.
• Strong knowledge of GPU efficiency and advanced AI architectures.
• Ability to develop and refine multimodal training systems.

Responsibilities

• Train LLMs / VLMs from scratch using distributed frameworks.
• Build and optimise multimodal training pipelines for text, image, and audio.
• Write and optimise CUDA / Triton kernels.

Skills

Large-scale GPU training

Triton / CUDA

Mixture-of-Experts architectures

Multimodal pipelines

Distributed frameworks

Tools

PyTorch

Megatron

DeepSpeed

Job description

Do you want to build frontier-level LLM models from scratch?

Have you worked on large-scale GPU training, Triton / CUDA, or MoE systems?

Are you ready to join one of Europe’s most technical deep-learning teams?

A Europe-based deep learning company is building the next generation of foundation models. Think of a smaller, faster, highly technical version of the major frontier labs – focused on LLM / VLM training, GPU efficiency, safety layers, and advanced architectures. They are preparing for their next funding milestone and operate with an extremely high technical bar.

They are hiring an AI Engineer to focus on training, scaling, and optimising large models. This role is hands‑on, research‑driven, and sits at the core of model creation. The AI Engineer will train LLMs and VLMs from scratch, optimise distributed GPU systems, and contribute to new architectures including Mixture‑of‑Experts and multimodal pipelines. You’ll work closely with a small team of world‑class engineers on one of the most technical problems in AI.

Key responsibilities

• Train LLMs / VLMs from scratch using distributed frameworks
• Build and optimise multimodal training pipelines (text, image, audio)
• Develop and refine Mixture‑of‑Experts architectures
• Write and optimise CUDA / Triton kernels
• Improve training stability, speed, and memory efficiency
• Experiment with new architectures, scaling laws, and data mixtures

Key details

• Salary : Up to £200k + equity (0.1–0.3%)
• Working model : UK, 100% remote
• Stack : PyTorch, Megatron, DeepSpeed, Triton / CUDA, multimodal architectures

Interested? Please apply below.