Data science interview questions in India: what actually gets asked
The lists of data science interview questions floating around India optimise for the wrong round. The definitions are the easy part and rarely where offers are won or lost. Here is what each round actually tests, from someone who has sat on the hiring side of the table, so you prepare for the interview you will get and not the quiz you have memorised.
Search for data science interview questions in India and you will find the same quiz everywhere: define p-value, explain the bias-variance tradeoff, what is a random forest. Memorise it and you will clear exactly one round, the easy one. The offers are won and lost in the rounds those lists do not prepare you for. I have interviewed for these roles, and here is what each round is actually testing.
The coding and SQL round
This is the filter, and in India it is more often SQL than people expect. A lot of “data science” work is pulling and shaping data, so companies check whether you can actually query. Practice joins, window functions, and group-bys until they are reflexive. If you freeze on a moderately nested SQL question, it does not matter how well you know gradient boosting, because you will not reach the round where it comes up. This is the cheapest round to prepare for and the most common one people underrate.
The concepts round, and the follow-up
Yes, they will ask you to explain overfitting and regularization. Know them cold. But the signal is in the follow-up, not the definition. After you define overfitting, expect “how would you detect it here,” and after you name a metric, expect “why that metric and not another for this problem.” The interviewer is checking whether you understand the concept or just memorised its sentence. Prepare by taking every definition you know and asking yourself the “so what, in practice” question behind it. That is the version that gets asked once you are past the fresher screen.
The project round, which decides it
They will pick something off your resume and go deep. This round decides more offers than any other, because it is the one you cannot cram for. If the project is genuinely yours, you can walk through why you made each choice and what you would do differently, and the conversation flows. If it was a tutorial you followed, you run out of answers in about three questions, and the interviewer knows. The preparation for this round happened months ago, when you either built something real or did not. If you have time before your interviews, that is where to spend it, which is the whole argument in how to actually become an ML engineer here.
The case or business round
Senior interviewers, especially at product companies, will hand you a vague problem and watch how you think. “Our numbers dropped, how would you investigate.” There is no memorised answer. They want to see you ask clarifying questions, break the problem down, and reason under uncertainty. You prepare for this by practising the thinking out loud, not by learning facts.
So prepare for the interview you will actually get. Get your SQL sharp, know your concepts well enough to survive the follow-up, and above all be able to defend a real project, because that is the round that separates the offers from the near-misses. The quiz is the floor. The rounds that decide it are about whether you can think, not whether you can recite.