Research Engineer (Materials Foundation Models)

present of AI; Cambridge; Hybrid; GBP 70,000 - 110,000; Full time

Job summary

CuspAI in Cambridge, UK seeks a scientist-Engineer to train and evaluate large AI models that learn from crystal structures, materials data, text and images. You own models from data to production, balancing experimental work with reliable, production-ready code.

The role involves multi-GPU distributed training, exploration of representation learning, LLM post-training or RL approaches, and collaboration with chemists to turn model outputs into lab decisions.

Qualifications

• Solid software engineering habits and hands-on experience with distributed training on many GPUs.
• Depth in representation learning or language model post-training or reinforcement learning.
• Ability to own models from data through to production integration.

Responsibilities

• Train and judge large AI models that learn from crystal structures, materials data, text and images.
• Balance experiments with production-quality code and robust evaluation.

Skills

Distributed training

Representation learning

LLM fine-tuning

RL post-training

Production-grade code

Job description

The role is about training and judging large AI models that learn from crystal structures, material properties, text and images. A strong candidate has solid software engineering habits, hands-on experience with distributed training on many GPUs, and real depth in one area such as representation learning, language model post-training or reinforcement learning. Each person owns a model from data through to integration, so comfort with both experiments and production quality code matters.

How it fits the AI era

Here AI is the product itself: the models propose and rank candidate materials, and engineers decide which modelling choices actually hold up through scaling studies and careful evaluation. What stays human is judging whether results make scientific sense, designing good data and tests, and working with chemists and materials scientists to turn model output into real lab decisions.

How it builds AI experience

The work involves training multi-modal foundation models on terabytes of scientific data across hundreds of GPUs, plus fine-tuning and post-training language models for scientific tasks. Someone new to materials can build a head start by studying how crystal structures are represented for machine learning and by trying a small project that learns embeddings from structured scientific data or figures.

Worth showing

• Distributed multi-GPU training project
• Self-supervised or contrastive learning
• LLM fine-tuning or RL post-training
• Evaluation suite and data curation
• Learning chemistry vocabulary quickly

This is our own read, written from the employer's posting and the company's Spotlight. The employer's posting is the source of truth for the role.

About CuspAI

An AI search engine for new materials that the company says turns a wanted set of physical properties into synthesizable chemical designs in months instead of years of lab trial and error.