Tell me about yourself for a data scientist interview

Say what the model or analysis was for. Accuracy on a notebook is not the result. The result is what a team did with it.

Build your answer

Target role The job you are interviewing for. Years of experience Years in this kind of work. Use 0 for a first role or a career change. Current role or background A field, a path, or a short phrase.

Win 1 Include a number. Win 2 Optional. Win 3 Used in the long version.

Key strength One habit, not five adjectives. Why this company or role A reason that would be wrong for a different employer.

Runs in your browser. Nothing you type is uploaded.

30-second version

About 34 seconds · 84 words

A short cut for a recruiter screen.

Stop after one result. A short screen does not need the second story.

60-second version

About 56 seconds · 141 words

The default. Present, then past, then future.

If they lean in, drop the second result and spend the time on how the first one happened. Leave duties from the job description out.

90-second version

About 87 seconds · 217 words

A longer cut when they ask you to walk through it.

I'm looking for a data scientist role. For the past 4 years I've worked as a data scientist on a search relevance team. Before that, this is the work that got me here. I shipped a ranker that lifted click-through 7% against the previous heuristic. Another result was killed a model that looked better offline and worse on a weekend slice. I then wrote the experiment readout 40 product managers still reuse. None of that is a title. It is the work I can walk through. The through-line is checking the slice where a pretty offline number is lying. The future I'm aiming at is a data scientist role, because this team ships models into a product, and I want that last mile. The first thing I'd measure is the constraint the team already feels. The numbers in those results are ones I can defend, including what I owned. The data scientist role is the move, and the results are how I know I can do the work. I can say any of those results more slowly, with the owner and the number intact. I would use the same judgment on the next data scientist problem. The work I want next is still data scientist work: a clear owner, a number, and a decision I can explain.

They can interrupt. Go present, then proof, then why this role. When they ask, go from the headline number to the decision you made.

When you can say the minute without the page, practice it live with Luna. It hears the question and helps you keep a line. 24 hours free.

How to answer it for this role

Data science openers drown in methods. Logistic regression, a ranking model, an experiment. Pick one and say the decision it supported, the baseline you beat, and whether it shipped. If it did not ship, say what you learned and do not dress it up as production.

Have the evaluation story one question deep: the metric, the slice where it failed, the leakage you checked. You do not tell that whole story in the minute. You tell enough that they believe you can.

The future beat should match their kind of science. Experimentation, forecasting, ranking, and research are different jobs. "I love machine learning" does not choose.

What to emphasize

Baseline and what changed.
Whether it shipped.
The decision owner.

What to leave out

A catalog of algorithms.
A Kaggle rank as your main professional proof.

Sample 60-second answer

This is the sample loaded into the builder above. About 56 seconds, 141 words. Replace every fact with yours before you practice it.

I'm looking for a data scientist role. For the past 4 years I've worked as a data scientist on a search relevance team. Before that, this is the work that got me here. I shipped a ranker that lifted click-through 7% against the previous heuristic. Another result was killed a model that looked better offline and worse on a weekend slice. The through-line is checking the slice where a pretty offline number is lying. The future I'm aiming at is a data scientist role, because this team ships models into a product, and I want that last mile. In the job I'd bring that strength to the team's current goal, and I'd report what moved. I have the detail under each number, including what I would measure again. That is the minute: the data scientist work, the proof, and the reason.

The 30-second cut is about 34 seconds. The longer cut is about 87 seconds. Both use the same facts in the same order. Read the longer structure guide if you want the timing rules in full: how long the answer should be.

Tips before you say it

Say the 60-second version out loud before you edit it. A sentence that looks fine on the page often runs long or sounds like a memo.
One number per win. “Improved the process” is a claim. “Cut hold time from 6 minutes to 2” is a result you can be questioned on.
Stay in order: present, past, future. If you start with where you grew up, the interviewer is waiting for you to arrive.
Leave a door open and then stop talking. The extra story you add in the silence is usually the one you did not prepare.
Keep your part separate from the team’s. “We launched” collapses the first follow-up. “I cut the scope, engineering shipped it” survives.

Other roles

Questions people ask

Can I mention papers or Kaggle?

A single clause, after a work result, if you are early-career. It should not be the only proof.

What if I mostly did analytics?

Interview for the job you did, or be explicit that you are moving toward modeling. Do not rename dashboards as machine learning.

How much math belongs in the opener?

None, unless the result is the math. The interview will test it. The opener tests judgment.