CommonAI CIC is a non-profit membership organisation, founded on a belief in collaborative engineering for the safe and responsible development of foundational AI technologies. A place where AI startups, enterprises large and small, public sector bodies and academia can share resources and knowledge, to codevelop and grow businesses, fast.

We are led by experienced founders, investors and engineers who believe that collaborative engineering drives faster AI innovation and are backed by a mix of UK Government and private funding in order to design, build and deploy innovative AI systems.

The Opportunity

We are seeking a Research Scientist to drive our research on new large language model architectures that reduce the cost of inference while retaining performance. The work treats the realities of serving hardware, such as memory bandwidth and data movement, as first-class design constraints. It runs on the UK's national AI supercomputers and on our own compute nodes.

You will join a small research team, working with our Research Software Engineer and kernel engineer, and take early-stage ideas through to rigorous experiments, larger-scale runs and publication.

As an AI-first company, we strongly believe in collaborative engineering powered by the tools we are helping to build. This role places a major emphasis on using LLMs, AI coding assistants, and autonomous agents for research and software development, and you will have a top-tier Claude Code subscription.

What You'll Do

• Design and implement new model components, such as attention variants, structured or sparse layers, routing mechanisms, activations and training objectives.
• Run controlled pretraining experiments at matched compute, from small scale up to larger models.
• Analyse the structure that trained models learn, and what it means for efficiency and interpretability.
• Explore adapting existing open models through continued training or distillation.
• Work with our engineers to turn research ideas into reliable, reproducible training runs and efficient kernels.
• Write papers, release code and present results.

Requirements

• A PhD in machine learning, computer science, physics or a related field, or equivalent research experience.
• Published or shipped research on new model components, such as attention mechanisms, activation functions, normalisation, mixture-of-experts routing, or structured or sparse layers.
• Hands-on experience pretraining transformer language models from scratch, including careful ablations at matched compute.
• Strong PyTorch (or deep JAX experience and willingness to work in PyTorch), including multi-GPU training.
• A working understanding of LLM inference costs, such as KV cache, memory bandwidth and why decoding is memory-bound.
• Rigour in experimental design: strong baselines, statistical comparison and honest reporting of negative results.
• A strong enthusiasm for using LLMs, coding assistants, and agents as core tools in your own research and development process.

We also value:

• GPU kernel work (e.g. Triton, CUDA) for sparse or custom attention.
• Experience with sparsity, pruning, mixture-of-experts or distillation of LLMs.
• Knowledge of accelerator and cluster hardware, including memory hierarchies and interconnects.
• Experience running workloads on HPC systems with a batch scheduler such as Slurm.
• Mechanistic interpretability experience.
• A quantitative background from another field, such as physics, applied maths or neuroscience.
• Interest in hardware–model co-design.

Benefits

• A collaborative and supportive work environment
• The opportunity to have a high impact in a growing organisation
• Competitive salary package and pension
• Professional development opportunities
• Networking opportunities with influential people from across the tech sector, financial services, and academia
• A vibrant office environment located a few minutes' walk away from Cambridge train station

CommonAI CIC is an equal opportunity employer and is committed to creating an inclusive and diverse workplace.