Machine Learning and AI Engineer Recruitment

A Founder told us recently that he had interviewed eleven Machine Learning Engineers and still could not say, honestly, which of them was any good. Budget signed off. Role open five months. Nobody in the building who could tell a strong ML engineer from a confident one.

That is the real problem with machine learning and AI hiring. Not supply. Judgement.

Runtime is a specialist Engineering Recruitment partner for SaaS and VC-backed technology companies. We hire Machine Learning Engineers, AI Engineers, MLOps and Applied Research people into teams that are shipping product, not publishing papers. Our UK headquarters is in Manchester and our US base is in Tampa.

Build_Ship_Runtime.

The Four Jobs hiding behind one Job Title

Most ML briefs that reach us describe at least two of these, sometimes all four.

Applied ML engineer. Takes a business problem, picks an approach, trains and evaluates a model, ships it. Comfortable with ambiguity in the problem definition itself.

AI or LLM engineer. Builds on top of foundation models. Retrieval, evaluation, orchestration, latency and cost. Increasingly this is a software engineering job with a research-shaped edge.

ML platform or MLOps engineer. Owns the pipelines, the feature store, the training infrastructure and the deployment path. Closer to platform engineering than to data science.

Research engineer. Reads papers, reproduces them, pushes a narrow area forward. Rare, expensive, and the wrong hire for most Series A and Series B products.

Getting this wrong is the most common reason an ML search stalls. We spend the first conversation working out which of the four you need, and we will tell you if we think the honest answer is a strong backend engineer who is curious about models.

Why Machine Learning searches stall

Three things, usually.

The brief is a wish list. Publications, distributed training, production experience, product instinct and a PhD. That person exists. They are not joining a twelve-person team, and holding out for them costs you six months.

The interview loop cannot tell signal from fluency. ML is an unusually easy field to sound good in. Without a structured way to test how someone reasons about data, evaluation and failure, panels default to whoever interviews most smoothly.

The pitch is generic. Strong ML people choose based on the data, the problem and who they will learn from. A job advert that says “cutting edge AI” tells them nothing. A paragraph about your actual dataset, your latency budget and the decision you are trying to automate tells them everything.

How we run a Machine Learning Search

We start by pinning down which of the four roles you are hiring, then we write the story a strong candidate would need to hear to leave a good job. That story is built from your problem, not from adjectives.

Sourcing is direct and mapped, not advert-led. For ML and AI roles the people worth hiring are rarely applying, and the market is small enough that a proper map of it is achievable rather than aspirational.

We screen on reasoning before credentials. What did you do when the model performed worse in production than in evaluation? How did you choose your metric? What did you decide not to build? Those answers separate people quickly.

We then hand you a shortlist with our reasoning attached, including the reservations. Eight CVs to a hire is our average, which means the shortlist is short on purpose.

Assessing ML Engineers when nobody in-house is one

This is the situation most of our clients are actually in, and it is workable. You do not need an ML Expert on the panel to run a credible loop. You need the right questions, a real artefact to review, and a clear view of what a good answer sounds like.

We have written up how to do this in practice: how to interview an ML engineer when nobody on your team is one. If it is useful, take it and run it yourself. If you would rather we sat on the panel, we do that too.

Where we Hire

UK and US, from Manchester and Tampa respectively, with coverage across Europe where clients are building distributed teams. Most of our machine learning work sits with SaaS companies between Seed and Series C, plus a smaller number of scale-ups building an AI capability inside an existing product organisation.

What our Numbers look like

Six weeks average placement velocity. Eight CVs per hire. 94% offer acceptance. 95% retention.

The retention number is the one we care about most on ML roles, because a mis-hire here is expensive twice over: you lose the year, and you inherit a system nobody else can maintain.

How we Engage

Container, our fixed-scope retained model for a defined set of roles. Executive Search for VP and Head of level hires. RPO where you need volume and process. Embedded, where one of our team works inside yours. Contingent where the brief is genuinely open.

Most ML and AI engineer searches run as Container or Embedded, because they need depth of market mapping rather than speed of CV flow.

Frequently asked questions

What does a Machine Learning Engineer actually do?

A Machine Learning Engineer turns a business problem into a model that runs in production. That means framing the problem, preparing and validating data, choosing and training an approach, deciding how success is measured, and then owning the model once it is live. In most SaaS companies it is closer to software engineering than to research, and the hardest part is usually evaluation rather than modelling.

What is the difference between a Machine Learning Engineer and an AI Engineer?

A machine learning engineer typically builds and trains models on your own data. An AI Engineer usually builds on top of existing foundation models, focusing on retrieval, prompting, evaluation, orchestration, latency and cost. The skills overlap but the day job is different, and hiring for one when you needed the other is the most common mis-hire in this area.

How long does it take to hire a Machine Learning or AI Engineer?

Runtime averages six weeks from brief to accepted offer across our engineering searches. Machine Learning roles sit at the longer end of that range when the brief is unclear at the start, and at the shorter end when the team has agreed in advance which of the four ML role types they are hiring and what the interview loop will test.

How do we assess ML Engineers if nobody on our team is one?

Use a real artefact rather than a whiteboard. Ask the candidate to walk through a model they shipped, then push on the parts that do not depend on ML expertise to judge: how they chose the metric, what the failure modes were, what happened when production performance diverged from evaluation, and what they decided not to build. A non-specialist panel can assess reasoning, honesty and judgement reliably. Bring in an external specialist for one technical session if you want depth on the modelling itself.

Do you recruit MLOps and ML Platform Engineers as well?

Yes. ML platform and MLOps engineers are a distinct market from applied ML, sitting closer to platform and infrastructure engineering. Many teams find that hiring a platform-shaped person first is what makes their applied ML hires productive, because the deployment path exists before the models do.

Which Locations and Company Stages do you cover?

Runtime recruits Engineering talent across the UK and the US, with a UK headquarters in Manchester and a US base in Tampa, plus coverage across Europe for distributed teams. Our Machine Learning and AI work is concentrated in SaaS and VC-backed technology companies between Seed and Series C, alongside later-stage scale-ups building AI capability inside an existing product organisation.

Ready to talk?

Tell us the problem you are trying to model and we will tell you which of the four roles you are actually hiring. Get in touch.