Meesho Data Scientist — Interview Guide

About Meesho

Meesho is a Bengaluru-headquartered Indian e-commerce marketplace built around small sellers and value-conscious shoppers. For a data scientist that means large volumes of transaction and browsing data and problems such as recommendations, search, pricing and fraud, which is why the interview leans on both engineering fundamentals and applied ML.

One thing is structurally unusual about how the process is reported. Candidates describe a first technical round that is not a pure data science screen. It can run from about 45 to 90 minutes and combine data structures and algorithms, SQL, statistics, ML basics and, in some reports, PySpark in a single sitting. Candidates who describe the process point to the DSA portion as the hardest part. Treat the first round as a generalist engineering screen with data science on top, not the other way around.

Interview Process Overview

Based on candidate reports from 2025, the process has roughly four stages: a recruiter screen or application, a screening assessment or first technical round, a deeper ML and resume-projects round, and, for more senior data scientist levels, a case study. Reports vary by level and team, so the number of rounds and their order are not fixed. Timelines are reported inconsistently, and some candidates say they heard nothing back after a later round, so do not assume a standard feedback loop.

These are candidate reports, not a Meesho statement. Treat the structure as a likely shape, not a guarantee.

Stage 1: Recruiter Screen and Online Assessment

Format: Entry is reported through recruiter outreach (one candidate was approached via a job portal) or a direct application. One candidate reports an online assessment covering statistics and SQL plus a single PySpark question.

What it tests: Baseline SQL, basic statistics and familiarity with distributed data processing.

What candidates report being tested on:

1. SQL at roughly medium difficulty, such as top-N per department, medians using a CTE or window function, and rolling sums.
2. Basic statistics, mostly fundamentals rather than advanced theory.
3. A PySpark question on how Spark executes work, such as lazy execution.

Stage 2: First Technical Round

Format: Reported lengths range from about 45 to 90 minutes. One candidate describes a single sitting that mixed coding, SQL, maths, resume discussion, ML and PySpark. Several candidates report not getting past this stage.

What it tests: Problem solving under time pressure across coding, SQL, statistics and ML basics.

Question types candidates report:

1. Medium-to-hard DSA problems. Reported examples include Combination Sum II, Asteroid Collision, a union-find graph problem and implementing logistic regression from scratch.
2. SQL: top-N per department, medians via a CTE or window function, rolling sums.
3. Statistics and ML basics: p-value versus power, Type I versus Type II errors, precision and recall, ROC-AUC, regularization, bias versus variance, bagging versus boosting.
4. Spark: lazy evaluation, UDFs, RDDs versus DataFrames, anti joins and semi joins.
5. One candidate reports being asked to compare LoRA and QLoRA, so recent fine-tuning methods can come up for roles with a generative AI angle.

Stage 3: ML Depth and Resume Projects

Format: Reported by two candidates. The interviewer works through your resume and goes deep on the projects you list.

What it tests: Whether you understand why you made each modelling decision, not only what you built.

Areas candidates report being probed on:

1. Algorithms and loss functions: why this loss, what it optimises, what its failure modes are.
2. Feature engineering: how features were created, validated and checked for leakage.
3. Decision logic behind each project: why this model over a simpler one, how you measured success.
4. Anything on your resume you cannot defend in detail. Candidates are advised to be ready to explain every project they list.

Stage 4: Case Study (reported at Data Scientist 2 level)

Format: One Data Scientist 2 candidate reports a case that began with the resume and moved into a recommendation-system problem.

What it tests: End-to-end applied thinking.

Topics reported:

1. Designing the recommendation pipeline from data to serving.
2. Choosing evaluation metrics, both offline and online.
3. Running and reading an A/B test for the change.

Candidates for other levels may not see this stage, and no separate HR or hiring-manager round is described in the reports we reviewed.

What Interviewers Look For

Based on the reported question mix, interviewers appear to value solid coding fundamentals, comfort with SQL at scale, a working grasp of statistics, the ability to explain modelling choices from first principles, and applied judgement on product problems such as recommendations. Breadth in a single sitting matters as much as depth in one area.

How to Stand Out

1. Practise DSA at medium-to-hard level before you touch ML revision, since that is the stage candidates say stops people.
2. Rehearse SQL window functions until top-N, medians and rolling sums are automatic.
3. Be able to state, for every project on your resume, the objective, the baseline, the metric and what you would change.
4. For the case study, practise a full loop: framing, data, model, offline metric, A/B test, and how you would read a flat result.

Use InterviewBee's AI mock interviewer to rehearse the mixed technical round under time pressure, then review the Data Scientist interview questions and answers to strengthen the SQL, statistics, machine learning and DSA fundamentals candidates report being tested on.

Common Mistakes to Avoid

1. Revising only ML theory and walking into a first round that is mostly coding and SQL.
2. Spending 60 minutes on one DSA problem in a long mixed round instead of budgeting time across sections.
3. Listing projects on your resume that you cannot explain at the loss-function and feature level.
4. Giving a recommendation-system answer that stops at the model and never reaches evaluation or A/B testing.