AI Scientist

Location: Hybrid (London, Moorgate)

Duration: Full-time, starting 1st August 2026

Many of the materials that will define the next century of energy and computation don't exist yet, so we're building the AI to find them. Mater-AI is an early-stage deep-tech startup in London using machine learning to discover novel thermoelectric materials. These materials can cool electronics with no moving parts and turn waste heat into usable electricity. The chemical space we're searching is too large to explore manually, so the models we build decide where to look.

About the Role

You'll build the engine that drives discovery. Our pipeline generates and screens candidate materials before we commit any to expensive first-principles simulation and lab validation. The screening models you build are the filters that determine which candidates are worth exploring thoroughly, and ultimately which materials we try to make real. Optimising these filters means breakthroughs faster than anyone in the field. Two problems sit at the heart of it: Models must be robust for extrapolation, not just interpolation; the interesting candidates live at the edges of known chemistry, exactly where the majority of models are most uncertain. Every first-principles calculation we run on a top candidate is a hard-won signal; you'll design how that data flows back into the models to improve future predictions, while guarding against mode collapse. This is a high-autonomy role. The models you choose and the data you trust will decide the proposed candidates, and what the company discovers. Rigorous evaluation and defensible decision-making are incredibly important.

What you'll do

• Design, build and further optimise the ML that screens generated candidates (ensembles, deep learning, chemistry-based whatever the problem demands), and ship your changes into a live pipeline.
• Push the pipeline to the frontier by bringing in new methods, agentic research workflows, and entirely new model types where necessary.
• Turn first-principles results into better predictions by building the feedback loop that improves the models that undertake generation and screening.
• Direct what gets discovered through your analysis of model outputs, which sets the materials that advance to simulation and the lab.
• Analyse model outputs and existing data to identify physically-meaningful patterns.
• Train and stress-test models at scale across HPC and VM GPU clusters, as well as on local machines.
• Get fluent in an existing scientific codebase quickly, then work to improve it.

Required

• Strong hands-on experience building and evaluating ML for scientific applications: deep learning, ensembles, generative models or related.
• Rigour in evaluation: appropriate statistical metrics, uncertainty estimation, reproducible benchmarking.
• Excellent software engineering (python) and a track record of version-controlled code.
• Experience orchestrating large-scale computational workflows on HPC and VMs, and running them efficiently.
• A knack for systems that detect, diagnose, and recover from failure with minimal supervision.
• Comfort reading and reasoning about a codebase you didn't write.
• Proven ability to work independently and deliver defensible innovative solutions in a fast-moving startup.
• Being proactive and a clear communicator across a multidisciplinary team of physicists, ML researchers, and founders.
• Right to work and live in the UK and fluent in English.

Nice to have

• Familiarity with computational materials science or first-principles workflows (DFT, MLIPs, MD, transport).
• Building agents for scientific discovery or engineering workflows (autonomous experimentation, code-writing agents, pipeline orchestration).
• A feel for data-sensitivity and IP: data residency, keeping proprietary data in trusted environments.
• Active learning, uncertainty quantification, reinforcement learning, or other adaptive techniques.
• HPC and GPU environments, containerisation, job schedulers.

Education

A formal background in a quantitative scientific or engineering discipline (PhD, postdoc, or MSc with 4+ years' experience), or equivalent proven industry experience building ML for science, ideally in materials discovery.

What we offer

• Real ownership: the models you build steer the whole discovery effort.
• Access to state-of-the-art GPUs, and problems that require their use.
• A founding team at the frontier of AI and computational materials science to build alongside.
• Competitive compensation and a flexible remote policy.