The candidate asking this question has already done something right: they are treating the interview as a practical problem to be solved, not a gauntlet to be survived. That instinct matters more than any single SQL function or pandas method. But the real signal here is the question itself. They want to know if they should be grinding LeetCode or polishing their data manipulation skills. The honest answer, based on the shape of the interview, is that they should be doing both, but not in equal measure.
The 60-minute format is a strong hint. A data science intern role at a company like Loblaw is not asking you to build a neural network from scratch or solve a dynamic programming problem that would make a quant sweat. It is asking you to demonstrate that you can work with data in a practical, reproducible way. SQL and pandas are the tools of that trade. General DS concepts are the framework. LeetCode-style algorithm questions have their place, but for an intern role, the emphasis is almost certainly on whether you can clean a messy dataset, join tables sensibly, and extract an answer that a business stakeholder would trust. That is not a lesser skill. It is the core skill.
So what should the candidate actually do? First, get comfortable with the common pandas operations: filtering, grouping, merging, and handling missing values. These are the building blocks. Then, practice writing SQL queries that mirror those same operations, because the logic transfers even if the syntax does not. For general DS concepts, focus on the fundamentals: bias-variance tradeoff, cross-validation, precision versus recall, and the difference between overfitting and underfitting. These are not trick questions. They are the vocabulary of the field. If you can explain those clearly, you are already ahead of a lot of people who memorized model architectures but cannot articulate why a baseline model matters.
The "vibe" of the technical screen, as the candidate puts it, is likely more collaborative than combative. Interviewers in this space are not trying to stump you. They are trying to see how you think. If you get stuck, talk through your reasoning. If you make a mistake, correct it out loud. That transparency is a feature, not a bug. And if you are unsure whether to prioritize LeetCode, err on the side of practical preparation. A candidate who can walk through a messy data problem and explain their steps is more memorable than one who can recite a textbook algorithm but freezes when asked to pivot a table.
Here is the concrete takeaway: spend the majority of your prep time on SQL and pandas, with a clear focus on the operations you would actually use in a business context. Spend the remaining time reviewing core DS concepts and doing a few medium-difficulty LeetCode problems just to keep your coding fundamentals sharp. The goal is not to be perfect. It is to be ready to show that you can turn raw data into insight, under time pressure, with someone watching. That is the job. The interview is just the first chance to prove it.