top of page
Human-first approach to deep tech recruitment and talent matching

Research scientist, research engineer or ML engineer: which hire do you actually need?

Nov 13
4 min read

How to turn your next technical milestone into a hiring brief, distinguish overlapping profiles and avoid asking


Hire a research scientist when your edge depends on a method nobody has published yet. Hire a research engineer when you have a promising method but cannot test it fast or reliably enough.

Hire an ML engineer when it already works in a notebook and now has to survive real traffic. Most Series A and B AI startups need a research engineer, not a pure research scientist, unless research output is the product itself.


The hard part is that the three titles overlap, and every company uses them differently. Here is how to choose without guessing.


Why do these three titles get confused?


Macro of a translucent snowflake or ice crystal on a dark background, shimmering white and silver with soft blurred edges

Because there is no standard. Chip Huyen's ML interview guide points out that at some labs there is little real difference between a research scientist and a research engineer, and that the title often reflects academic status and pay band more than the work. She describes a candidate who picked a startup over a big-name company simply because the startup offered the research scientist title.


Companies also run different naming conventions. ISG Partners observes that in one company "machine learning engineer" means an applied modeller who ships, while in another every AI role carries that title. So a job title tells you little.


The work the person will own tells you everything. That is the rule for the rest of this article.


What does each role actually own?


Research scientist

Research engineer

ML engineer

Owns

Problems with no published solution

The systems, data and experiments that make research testable

Systems that train, serve and monitor models in production

Output

New knowledge: methods, papers, novel results

Reproducible experiments and tooling, often co-authored results

Reliable, scalable, monitored production systems

Typical background

PhD and strong publications, usually

Strong software engineering plus ML; PhD not required

Software engineering plus ML deployment

Hire when

Your defensibility depends on solving what established methods cannot

Ideas exist but experiments are slow, manual or not reproducible

Models work in notebooks but fail under real load

The research engineer is the one most often misread. A hiring guide published through the Computing Research Association puts it plainly: it is an engineering hire, not a research scientist hire, and the person is measured on working, reproducible systems rather than papers.


How do you turn a technical milestone into a hiring brief?


Start from the milestone, not the title. Ask four questions, in this order.

  1. What is actually blocking the milestone? An unknown method points to a research scientist. A known method you cannot test quickly points to a research engineer. A model that breaks in production points to an ML engineer.

  2. Is success a result or a system? "Show that approach X beats our baseline" is a research outcome. "Run 50 experiments a week without manual work" is an engineering outcome.

  3. What should exist after 90 days? A written hypothesis and first results, a working experiment harness, or a monitored service. If you cannot name it, the brief is not ready.

  4. What is already in place? Data pipelines, evaluation and compute. Build the harness before hiring the person who builds the model.

Three examples of how a milestone becomes a brief:

  • "Our domain model needs to handle a task no benchmark covers." Method risk is high. Start with a research scientist or a senior research engineer with domain depth.

  • "We have three promising approaches and can only test one a month." This is an infrastructure problem. Hire a research engineer.

  • "The prototype impressed investors but latency and cost explode at scale." Hire an ML engineer with production depth.


What goes wrong when one person covers three jobs?


The unicorn brief is the most common mistake we see: a PhD with top publications, who also builds the infrastructure, ships production systems and leads the team. Candidates like that are rare, and the people who do apply tend to be strong in one area and stretched in the others.


The mismatch also does damage once someone is hired. One analysis warns that putting a research scientist on uptime and GPU utilisation targets chokes off the exploration you hired them for. ISG Partners describes the reverse: an engineer placed in a research seat can produce a wrapper around somebody else's paper and call it a result. And a researcher placed in an applied seat tends to produce careful work against a fast metric, then leave for a lab.


How should you screen each profile differently?


Use a different test for each. One practical framing: give a research scientist an open technical question and look at their experimental rigour, their baselines and what evidence would change their mind. Give a research engineer an open problem and ask what system they would build to investigate it. Give an ML engineer a production system and probe data, deployment, observability and cost.


If your interview loop is the same for all three, you are measuring the wrong thing for at least two of them.


Frequently asked questions


Do I need a PhD for a research role? For industry research scientist roles, a PhD and publications are the norm. For research engineers, they are generally not required. Strong engineering with real ML experience matters more.


Can an ML engineer do research work? Sometimes, if the problem is well defined and the method exists. If the right approach is unknown and the work means designing new experiments or architectures, you need a research engineer or scientist.


Which should I hire first? Hire for your current blocker. For many early AI teams that is a research engineer or an applied ML engineer. A research scientist comes earlier only when original research is central to your advantage.


Not sure which profile your milestone needs?

Send us your next technical milestone in three sentences. We will tell you which profile it points to, whether the market has that person, and what the brief should ask for.



Sources

  • Computing Research Association career center, "How to Hire a Research Engineer: A Complete Guide for 2026"

  • Chip Huyen, "Machine Learning Interviews" (research scientist vs research engineer)

  • ISG Partners, "AI Research Scientist vs ML Engineer vs Applied Scientist"

  • MyDataWorks, "AI Engineer vs ML Engineer vs Research Scientist"

  • Harmix, "AI Researcher vs. AI Engineer vs. ML Engineer Explained"

Comments


bottom of page