Machine learning engineer or data scientist: which role fits your evidence

The two roles start from the same model and ask for different proof of what happened next. A data scientist CV needs to show what the work decided or changed. A machine learning engineer CV needs to show that the work runs. Sorting your own projects by which of those you can already evidence is a better guide to the right role than the title you would like to hold.

The rest of this piece covers what each CV has to show, a way to sort your own projects, an invented example, what to do when a posting blurs the line, and what to do when your evidence leans the other way from the role you want.

Same starting point, different proof

Both roles begin with a model: data cleaned, a method chosen, a model trained and checked. That part looks much the same on a CV, which is why candidates and postings mix the two titles up.

The difference is in the sentence that follows the model. In one it is about a decision. In the other it is about a system. This piece says nothing about which role is harder or better paid. It is only about what kind of proof each one reads for.

What a data scientist CV has to show

A data scientist CV needs the problem someone wanted answered, the method chosen, how the result was checked, and what it decided or changed. A model that ends at a good score reads as unfinished, because the reader cannot tell whether it mattered.

The outcome can be small. An analysis that someone used to choose between two options is an outcome. For some postings a model that shipped is itself the outcome, so the balance shifts with what the posting asks. The data scientist CV checker sets out what these postings tend to read for.

What a machine learning engineer CV has to show

A machine learning engineer CV needs to show that the model runs. It might be served through an API, scheduled as a job or embedded in something else, and it is watched for failure and built with ordinary engineering practice such as version control and tests.

A CV that stops at training reads as research, because nothing shows the model ever ran outside a notebook. The machine learning engineer CV checker covers the checks for these postings.

Sort your own evidence

For each project, ask one question: what happened after the model produced an output? There are three honest answers.

• It informed a decision or an analysis. Someone looked at the output and chose something differently.
• It ran as part of a system. The output went somewhere automatically, and something watched whether it kept working.
• Nothing followed. The model was trained and scored, and the project ended there.

Projects in the first group are data scientist evidence. Projects in the second are machine learning engineer evidence. The third group supports neither role yet, and it is useful to know which projects are in it before you choose a target.

None of this needs a job title in either role. A project counts on the specific work it shows, and Do projects count without a job title? makes that case in full.

A sorted example

Here are three invented CV bullets, written for this article and not taken from any real CV. None of them states an accuracy figure. That is deliberate, because a number would not change which role the evidence supports.

Leans data scientist: "Built a demand forecast for a small shop and walked its owner through the results, who used them to change how much stock to order for slow weeks."

Leans machine learning engineer: "Wrapped a model that estimates a building's next day electricity use in a prediction endpoint, retrained it on a schedule, and added a health check that raises an alert when predictions stop arriving."

Has both halves: "Trained a classifier that sorts maintenance requests for a student housing association, gave the housing team a ranked list they used to decide which requests to check first, and ran it as an hourly job that logs its own failures."

The first says nothing about how the forecast was run, which is fine for a data scientist posting. The second says nothing about what anyone did with the predictions, which is fine for an engineering posting. The third shows both, so it can be described honestly to either role by bringing forward the half that posting asks for.

When the posting blurs the line

Postings do not always sort cleanly. Read the responsibilities rather than the title. If they are about decisions, analyses and the people who use the results, the posting reads for data scientist proof. If they are about serving, monitoring and pipelines, it reads for engineering proof.

Some postings genuinely ask for both, and a project with both halves, like the third bullet above, is the strongest fit for them. This is a way to read one posting, not a claim about how any employer hires. If the posting says AI engineer instead, the checks differ again, and what recruiters look for on an AI engineer CV covers them.

If your evidence leans one way and you want the other

Never close the gap by inventing. There are two honest moves instead.

The first is to describe the half you did but never wrote down. A script that served predictions to a colleague, or a chart that answered a real question, is evidence you already have. The second is to name what is genuinely missing. If there is no deployment and the role needs one, a small real addition, such as serving a model you have already trained, closes the gap, and the CV can say so plainly.

A five minute decision

Write one line for each of your projects and answer the same question for each: what happened after the model produced an output?

• Most in the first group: target data scientist postings.
• Most in the second group: target machine learning engineer postings.
• Most in the third group: pick the role you want, then add the missing half for real.

Once you know which role your evidence supports, MyRecruiterCheck can compare your CV against one specific posting for it.