Python is the single most listed skill among candidates on MyInternships.in, at 21,874 profiles. Scikit-learn, the library most beginners use to train their first model in Python, is listed by 406. Somewhere between "I know Python" and "I have trained and evaluated a model", most profiles stop naming things.
The figures are drawn from 55,557 profiles on our platform that list skills, as counted on 1 October 2026. They describe what candidates list, not measured competence, and should not be read as a national picture.
A data science intern's week, without the gloss
The popular image is of building clever models. The reality for an intern is mostly preparing data so that a model, or a simple analysis, becomes possible.
You may be asked to pull a year of customer records, work out why a third of one column is empty, decide what to do about it, and justify the choice. You may build a baseline model and find that a plain rule does nearly as well. Reporting that honestly is part of the job.
Interns also run experiments designed by seniors, compare results, tidy notebooks into reusable scripts, and present findings to people who do not care about the algorithm and want to know what to do differently on Monday.
The two roles overlap heavily at intern level. An analyst explains what happened; a data scientist tries to estimate what will happen or to automate a decision. In practice, a data science intern needs everything an analyst needs, plus enough statistics and programming to build and test a predictive model.
What gets tested
- Python fluency: writing a clean function, not just running notebook cells in order.
- Pandas and NumPy: filtering, grouping, merging and reshaping without looking everything up.
- SQL: because the data lives in a database before it reaches your notebook.
- Statistics: distributions, hypothesis tests, confidence, and why a result might be chance.
- Modelling basics: train and test splits, overfitting, choosing a metric that suits the problem.
- Explanation: describing a model's result to a non-technical listener.
A common interview exercise hands you a small dataset and asks what you would check before modelling anything. The right answer starts with questions about the data, not with an algorithm.
The count falls at every step
| Skill listed | India | Bengaluru | Hyderabad | Pune | Chennai | Delhi | Mumbai |
|---|---|---|---|---|---|---|---|
| Python | 21,874 | 2,458 | 1,968 | 1,778 | 1,550 | 1,162 | 899 |
| SQL | 21,804 | 2,440 | 1,974 | 1,933 | 1,487 | 1,223 | 891 |
| NumPy | 8,496 | 914 | 767 | 782 | 565 | 496 | 362 |
| Pandas | 8,056 | 861 | 751 | 736 | 548 | 485 | 359 |
| Machine Learning | 3,281 | 344 | 199 | 229 | 208 | 154 | 152 |
| Statistics | 578 | 70 | 37 | 58 | 18 | 45 | 40 |
| Scikit-learn | 406 | 48 | 26 | 31 | 24 | 22 | 18 |
Python and SQL are level, each near twenty-two thousand. NumPy and Pandas come next, listed by 8,496 and 8,056: well under half of the Python profiles name the two libraries that data work in Python depends on.
Machine Learning appears on 3,281 profiles. Then two small figures: Statistics at 578 and Scikit-learn at 406.
Some of this is labelling. A candidate who lists Machine Learning has probably used scikit-learn and did not think to name it. But the Statistics figure is harder to explain away. Roughly 38 profiles list Python for every one that lists Statistics. If you have real statistical grounding, from a degree or from disciplined self-study, state it clearly, because it is uncommon among profiles here.
A project with a question, not just a model
Avoid the datasets every beginner uses; reviewers have seen them many times. Choose data connected to something you understand, then:
- State one question that someone would actually pay attention to.
- Explore and clean the data, recording each decision and why you made it.
- Build a simple baseline first, then one better model, and compare them fairly.
- Report the metric, and say where the model fails and for whom.
- Finish with a short recommendation in plain language.
Publish the notebook and a tidy script. The comparison with a baseline is what most student projects skip, and it is what shows you understand evaluation. List it in the projects section of the resume builder.
Listings that deserve caution
Be clear-eyed about roles advertised as data science that consist of labelling images or copying records between sheets. Annotation is real work, but it should be described as such.
Step back from any offer that requires you to buy a course, pay for "placement support", or complete a long unpaid assignment using what looks like a company's live data. A fair screening task is short and uses sample data.
A genuine listing tells you the kind of data, the tools, the team you join and whether a mentor reviews your work. If none of that is present, ask before you accept.
Making the application count
Tailor the top of your résumé to each listing. If the role mentions forecasting, lead with anything time-related you have done. Quantify with care: describe the size of the dataset and the improvement over the baseline, not inflated accuracy claims.
In your note, link the project and say in one sentence what it found. Keep a second, different project in reserve to talk about at interview.
Research-flavoured roles also appear at smaller firms, so include startup internships, and scan fresher jobs if you are graduating soon.
When your project has a baseline, a comparison and an honest limitations note, browse current data science internships and apply to those whose data problems you find interesting.


