What you’ll be doing

On most projects the first job is finding out: what’s been tried, what actually works, and what’s genuinely open. Depending on the project mix:

• Writing production-quality Python the rest of the team builds on: clean interfaces, tests, pipelines that run the same way twice.
• Analysing an approach before building it: reasoning out how a calibration scheme, a reward function or a model should behave, and developing from that.
• Building models, simulators and synthetic-data generators for physics, sensors and engagement mechanics, because most of the data we need can’t be collected. So we make it.
• Training supervised, self-supervised and reinforcement-learning models on data you’ve first had to fight into shape: scientific archives, sensor feeds and operational records that arrive nothing like a benchmark dataset.
• Working out why a run stalled – a data defect, a broken reward, an unstable representation – and fixing it.
• Getting models out of the simulator and onto hardware: software-in-the-loop flight stacks, bench integration, single-board computers.
• Engineering the systems around language models: retrieval that surfaces the right evidence, outputs whose structure you can guarantee, evaluation that catches failures before the customer does, deployment into fully disconnected environments. Almost none of this is prompting.
• Talking to customers and specialist partners directly: understanding the operational problem, presenting what you found (including when the answer is “this doesn’t work”), and shaping what happens next.
• Owning a workstream: turning a vague question into concrete next steps, flagging risks early, and pulling in the team when something surprising turns up.

Who we’re looking for

Nobody has all of this. Strong candidates usually have most of the first list and depth somewhere in the second:

• Strong Python and real software engineering: you can design something others build on, and debug code you didn’t write.
• You’ve applied ML for real at least once (in a job, a serious internship, or an ambitious project of your own) and made it work – not just training runs, but the data, evaluation and integration around them. Courses alone aren’t enough.
• The instinct to investigate: form a hypothesis, test it, read the result honestly, change course.
• Clear writing and speaking: customers, reports and our own records depend on it.
• Comfort owning things in a small team: proposing next steps on fuzzy problems without waiting for a full spec.
• A master’s in a quantitative field, or work that shows we shouldn’t care – we hire on what you can do.
• Able to meet UK baseline security screening (BPSS) and our government customers’ pre-project checks (see the note below).

Depth in at least one of:

• reinforcement learning, ideally multi-agent
• self-supervised learning
• simulation and synthetic data
• mathematical and physics-based modelling of real systems
• making ML run within tight compute budgets – efficient inference on embedded and single-board hardware
• engineering production systems around language models (retrieval, structured generation, evaluation, locked-down deployment) – as systems work, not prompting
• graph neural networks
• probabilistic modelling and knowing when a model’s confidence can be trusted.

A note on vetting: You don’t need security clearance to join us. We do run standard baseline screening (BPSS), and our government customers make their own checks before anyone joins their project. Those checks are routinely fine for UK nationals and usually for nationals of allied countries. If you’re not sure where you’d stand, ask us – we’d rather have the conversation than have you rule yourself out.

What we can offer you

Most of what we can offer is the work itself: people here typically move across several of those projects in a year rather than settling into one corner. On development: there’s no formal training budget. What there is instead is the fastest learning we know – hard problems, breadth and experienced people at close range, the kind you learn from by working alongside them.

Some of our benefits:

• Flexible working. People have different responsibilities and interests, and a rigid working day suits few of them. We trust you to organise your own work.
• Remote-first, with a weekly in-person day in London (travel and lunch on the company) for those who can get there. Most of us treat it as the highlight of the week, and regular socials besides.
• 25 days’ holiday plus bank holidays, rising by a day each year to 30.
• Pension: 8% employer contribution when you contribute at least 5%; 5% employer contribution otherwise.
• Life insurance and income protection, employer-funded, as standard.
• Private health insurance and critical illness cover, premiums paid by us. These are opt-in only because HMRC taxes them as benefits in kind, so whether the cover is worth the tax is your choice.
• Professional membership of a relevant qualifying body, paid by us.

If this sounds like your kind of work, please apply – and make sure your CV, or something it links to, shows us one thing you’ve built or figured out. That’s what we look for first. Shortlisted applications are read by the engineers you’d work with.