AI research internships at industry labs are mostly aimed at PhD students, and some accept strong master's or undergraduate candidates with publications. The interview is less about trivia and more about whether you can do research: explain your own papers deeply, reason about math and experiments, and write working code. Candidate write-ups commonly describe a recruiter screen, a coding interview and a research interview, often followed by conversations with possible host researchers. This page covers the rounds to expect, the questions that come up with what a strong answer contains, two sample answers, what interviewers expect from an intern, the mistakes that cost offers, and a two-week prep plan.
What the AI research intern interview looks like
| Round | What happens | How to prepare |
|---|---|---|
| Recruiter screen | Your research area, publications, timing and which teams fit | A two-minute summary of your research agenda |
| Coding interview | A standard algorithms problem or an ML coding task such as implementing a model component | Write clean Python and NumPy or PyTorch without autocomplete |
| Research interview or talk | A deep dive into one paper, or a short presentation with questions | Prepare to defend every design choice, baseline and ablation |
| Host matching | Calls with researchers who might supervise you, discussing a possible project | Read their recent papers and bring one concrete idea |
Questions that come up, and what strong answers contain
- Walk me through your best paper. The problem, why it matters, the key idea, the experiments that support it, the strongest baseline, the limitations and what you would do next.
- Why does self-attention scale quadratically with sequence length? Every token attends to every other, so the score matrix is n by n; compute and memory grow with n squared, which motivates sparse, linear or chunked alternatives.
- Your model improved by one point on a benchmark. How do you know it is real? Multiple seeds, variance or confidence intervals, a fair tuning budget for baselines, checks for test-set leakage and ablations.
- Implement multi-head attention, or a training loop, from scratch. Correct shapes, masking, numerical stability such as subtracting the max before softmax, and a quick test.
- Derive the gradient of softmax cross-entropy. The gradient with respect to the logits is the predicted probabilities minus the one-hot target, with the reasoning.
- Pick a recent paper outside your area and critique it. What is new, what is convincing, what is missing and an experiment you would add.
- What would you work on during the internship? A scoped project that fits the team, a first experiment you could run in week one and a realistic result for twelve weeks.
Two sample answers
Your model improved by one point. How do you know it's real?
First I'd check variance: run at least three to five seeds for both my method and the baseline and report the mean and spread. If the gap is within the seed noise, it isn't a result yet. I'd make sure the baseline got the same tuning budget, confirm there's no overlap between training data and the test set, and run an ablation that removes my change. Finally I'd test on a second dataset, because a gain that only shows up on one benchmark is often overfitting to it.
What would you work on during the internship?
I've read your group's work on long-context retrieval, and my thesis is on efficient attention, so I'd propose testing whether my chunked attention method keeps retrieval accuracy at longer contexts than your current setup. In week one I'd reproduce your baseline on a small model; by week four I'd want a first comparison; and the realistic twelve-week goal is a clear positive or negative result with ablations, written up as a workshop paper or an internal report.
What interviewers expect from an intern
- Deep, honest knowledge of your own work, including what did not work.
- Experimental rigor: seeds, baselines, ablations and leakage checks.
- Math fluency in linear algebra, probability and optimization at the level of your field.
- Coding that runs, since interns are expected to run experiments from week one.
For the full-time version of these questions, see machine learning engineer interview questions.
Mistakes that cost offers
- Overclaiming results or hiding a weak baseline.
- Knowing your paper's headline but not its implementation details.
- Treating the coding round as optional because you are a researcher.
- Coming to host calls without having read the host's work.
A two-week prep plan
- Days 1 to 3: prepare a fifteen-minute talk on your best paper and a two-minute version.
- Days 4 to 6: coding practice, alternating standard algorithm problems with implementing model components from scratch.
- Days 7 to 8: math refresh: matrix calculus, probability and optimization for your area.
- Days 9 to 11: read two recent papers from each target team and write one-paragraph critiques.
- Days 12 to 14: mock research interviews and a mock coding round; our page on AI and ML internships at big tech covers the wider landscape.
Timing matters as much as preparation: see our internship recruiting timeline, and once you are in, turning an internship into a return offer. Engineering candidates should also read software engineer intern interview questions.
Where CoPilot Interview fits
Intern interviews reward clear explanations of your own projects and coursework, and that takes practice out loud. CoPilot Interview is an AI interview assistant; its free AI mock interview builds AI research intern questions from your résumé and the job description, reads them aloud, scores each spoken answer out of ten and shows a model answer to compare. The free plan gives 15 minutes a day of live help in sessions of up to 15 minutes, with AI mock interviews included, with no card, which suits a student budget. For real interviews and online assessments, follow your employer's rules.
Disclosure: CoPilot Interview is our product, so weigh what we say about it accordingly.
Frequently asked questions
Do I need publications for an AI research internship?
For most industry lab research internships, publications or strong preprints are expected, and many postings ask for PhD enrollment. Applied ML internships are more open to candidates without papers.
Is there LeetCode-style coding in research intern interviews?
Often there is at least one coding round, as candidates commonly report; the difficulty and style vary by lab and team, so ask the recruiter.
When should I apply?
Many labs hire in the autumn and winter for the following summer, but timing varies by lab and year; check each careers page and ask your advisor's network.
Can I use AI tools during an internship interview?
Only where the instructions allow it. Online assessments and many live rounds forbid outside help. Use AI to practice beforehand, and follow your employer's rules.
Practice this interview before the real one
The AI mock interview builds questions from your resume and the job description, reads them aloud and scores each spoken answer out of 10. Included on the free plan.
Start a mock interview