Company interview guide

Swiggy Machine Learning Interview Guide (2026)

A broad data science and ML loop: product thinking, SQL and pandas, statistics, then a deep LLM and project round.

Swiggy's machine learning interview loop has 5 reported stages (Product thinking, Coding: SQL, Python and pandas, ML breadth: statistics and classical ML, ML depth: LLMs, transformers and your projects, Hiring manager); typical length: Not reported.

Roles covered: Data Scientist and ML Engineer. The question bank holds 12 questions reported from Swiggy interviews, 3 of them free to solve in your browser.

Try a free question

Interview process length
Not reported
Stages
5
Reports linked to bank questions
13
Bank report dates
Aug 2026–Aug 2026
Questions in bank
12 (3 free)
Last reviewed
October 2026

How the Swiggy interview loop works

This outline rests on only two first-hand reports, so read it as a likely shape rather than a fixed loop. The fullest account (a 2026 Data Scientist 2 candidate) had five rounds in this order: product thinking on a scenario drawn from the candidate's own project, live coding with one SQL, one Python and one pandas question, ML breadth on statistics and classical ML, ML depth on LLMs, transformer math and resume projects, and a hiring-manager round. A 2023 NLP engineer report describes interviews with several panel members from different domains, questions running from basic to advanced, interest in system design, and real-world scenarios at the end. Neither report gives round lengths or an overall timeline, and we found no official description of the process.

The Swiggy interview process, stage by stage

01Product thinking

One round on a product scenario, in one report built from the candidate's own project; length not reported

The 2026 candidate was asked how an automated, LLM-based check from their own work should be presented to the sellers it affects: what they see while it runs, on success and on failure. The point is to reason from the user's and the business's side, not only the model's. The older NLP report also mentions real-world scenarios near the end of the panel, though without detail.

What it tests

  • Thinking about the end user of an ML system, not just its accuracy
  • Handling latency, uncertainty and failure states in the product experience
  • Explaining model output to a non-technical user
  • Connecting a past project to a business outcome

How to prepare

  • Walk through the happy path, the failure path and what the user feels at each step.
  • Pick projects on your resume that touched real users; expect the scenario to come from one of them.
  • Set expectations explicitly: what wait time to show, and who follows up when the model is unsure.

Opening round in one 2026 data scientist report; a 2023 NLP engineer report mentions real-world scenarios without saying whether they formed a separate round.

02Coding: SQL, Python and pandas

Live coding, three short questions (SQL, Python, pandas) in one report; length not reported

In the 2026 data scientist loop this round had one question in each of SQL, Python and pandas, all at a practical rather than puzzle level: a per-partition total next to every row with a window function, a classic single-pass array problem, and a per-customer rank in pandas. The SQL table was delivery-partner earnings, so expect food-delivery flavoured data. We have no report for an ML engineer coding round, which may lean more on algorithms.

What it tests

  • SQL window functions such as SUM and RANK with PARTITION BY
  • pandas groupby with rank, transform and rolling windows
  • Simple one-pass array logic in Python with stated complexity
  • Writing correct code quickly on small, concrete tables

How to prepare

  • Know the difference between a window aggregate and GROUP BY, and when each keeps every row.
  • Practise pandas rank methods (dense, min, first) and descending order per group.
  • State time and space complexity for the Python question without being asked.

Described only in one 2026 data scientist report, which gives the three questions; no other report covers a coding round.

03ML breadth: statistics and classical ML

One round of rapid conceptual questions on statistics, regression and classical ML; length not reported

The 2026 candidate calls this the round where weak fundamentals show, and it is the round they did not clear. Questions covered the central limit theorem, Bayes' rule, A/B testing and p-values, bias and variance, the assumptions of linear regression and what breaks when each fails, and why CNNs beat plain fully connected networks on images. The 2023 NLP report also describes questions moving from basic definitions to advanced topics.

What it tests

  • Central limit theorem, conditional probability and Bayes' rule
  • Hypothesis tests, p-values and the pitfalls of A/B testing
  • Linear regression assumptions, diagnostics and fixes
  • Bias-variance trade-off and regularisation
  • Why convolutional networks suit images

How to prepare

  • For each regression assumption, be ready with how to detect a violation and how to fix it.
  • Explain statistics in plain words first, then give the formula.
  • Do not skip the maths: this is where one candidate says the loop ended for them.

Free reported questions for this stage

A full statistics and classical ML round in one 2026 data scientist report; a 2023 NLP engineer report describes questions from basic to advanced, starting with definitions.

04ML depth: LLMs, transformers and your projects

One deep round on LLM choices, transformer maths and resume projects; system design may come up; length not reported

Expect to defend every decision in your projects: whether the training data matched production, why you prompted by hand or automatically, when to fine-tune instead of prompting, and how you chose an LLM by task, cost and latency. The 2026 report adds attention from first principles, the reason for scaling by the square root of the key dimension, how transformers handle images, and object detection with YOLO because it was on the resume. The 2023 NLP report says interviewers were also interested in system design.

What it tests

  • Attention, the roles of Q, K and V, and the scaling factor
  • Fine-tuning vs prompting vs prompt tuning, and choosing a model
  • Data representativeness and evaluation in your own projects
  • Vision transformers and object detection, when on your resume
  • Talking through an ML system design

How to prepare

  • Derive attention with tensor shapes on paper, and explain what goes wrong without the scaling.
  • Anything on your resume is fair game in full depth, including model internals and loss functions.
  • Have a clear rule of thumb for model choice: task fit, then cost, then latency.

Free reported questions for this stage

A detailed LLM and project depth round in one 2026 data scientist report; a 2023 NLP engineer report mentions advanced topics and system design without round detail.

05Hiring manager

Final conversation with the hiring manager; length not reported

The 2026 report lists culture fit, past experience, teamwork, conflict resolution and prioritisation as the focus. That candidate stopped after round four, so this summary does not come from someone who sat the round.

What it tests

  • Teamwork and handling conflict
  • Prioritising competing work
  • Depth and ownership in past roles
  • Fit with the team and the company

How to prepare

  • Prepare two or three stories on conflict and prioritisation with clear outcomes.
  • Ask what the team owns and how data science work reaches production there.

Named as the fifth round in one 2026 data scientist report whose candidate did not reach it; no other report mentions it.

Questions and themes reported most often

  • Explain attention, Q, K and V, and why scores are scaled by the square root of the dimension (1 reports, ml depth: llms, transformers and your projects)
  • Linear regression assumptions and what breaks when each one fails (1 reports, ml breadth: statistics and classical ml)
  • Fine-tuning vs prompting, and how to pick an LLM for a task (1 reports, ml depth: llms, transformers and your projects)
  • Central limit theorem, p-values and running an A/B test (1 reports, ml breadth: statistics and classical ml)
  • Show an automated ML check to the end user, including failure states (1 reports, product thinking)
  • SQL window total per partition next to every row (1 reports, coding: sql, python and pandas)
  • Rank or aggregate within each group in pandas (1 reports, coding: sql, python and pandas)
  • Best single buy and sell from a price list (1 reports, coding: sql, python and pandas)

What Swiggy asks about

  • LLMs, transformers & deep learning30%
  • Statistics & classical ML25%
  • SQL, Python & pandas20%
  • Product thinking15%
  • Hiring manager & behavioral10%

Swiggy interview prep checklist

  • Can list the linear regression assumptions with a diagnostic and a fix for each
  • Can explain the central limit theorem, Bayes' rule and an A/B test in plain words
  • Wrote a window-function query and a pandas per-group rank without looking anything up
  • Derived scaled dot-product attention with tensor shapes
  • Have a clear answer on fine-tuning vs prompting and on choosing an LLM
  • Rehearsed a product walk-through of one project, including failure states
  • Asked the recruiter for the type and length of every round

Tips

  • Confirm the round list with your recruiter early; our outline comes from very few reports.
  • Revise statistics and classical ML as seriously as LLMs; the breadth round is where one candidate was stopped.
  • Drill SQL window functions and pandas groupby, rank and transform until they are automatic.
  • Be able to derive attention with shapes and explain the scaling factor without notes.
  • Prepare every project on your resume in depth: data, evaluation, model choice and what you would change.
  • For product questions, cover the happy path, the failure path and the user's experience at each step.
  • Expect examples in the food-delivery domain, such as delivery-partner earnings or seller onboarding.

See the rest of the Swiggy loop

The later stages, the questions reported for them and how to prepare are in Pro.

Unlock with Pro

Practice the Swiggy loop

Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.

Swiggy interview FAQ

How many interview rounds does Swiggy have for ML roles?

Candidates report 5 stages: Product thinking, Coding: SQL, Python and pandas, ML breadth: statistics and classical ML, ML depth: LLMs, transformers and your projects and Hiring manager. This outline rests on only two first-hand reports, so read it as a likely shape rather than a fixed loop. The fullest account (a 2026 Data Scientist 2 candidate) had five rounds in this order: product thinking on a scenario drawn from the candidate's own project, live coding with one SQL, one Python and one pandas question, ML breadth on statistics and classical ML, ML depth on LLMs, transformer math and resume projects, and a hiring-manager round. A 2023 NLP engineer report describes interviews with several panel members from different domains, questions running from basic to advanced, interest in system design, and real-world scenarios at the end. Neither report gives round lengths or an overall timeline, and we found no official description of the process.

How long does the Swiggy ML interview process take?

Reported timelines run Not reported.

What does Swiggy ask in ML interviews?

By reported question mix: LLMs, transformers & deep learning 30%; Statistics & classical ML 25%; SQL, Python & pandas 20%; Product thinking 15%; Hiring manager & behavioral 10%. The most repeated themes: Explain attention, Q, K and V, and why scores are scaled by the square root of the dimension; Linear regression assumptions and what breaks when each one fails; Fine-tuning vs prompting, and how to pick an LLM for a task.

How should I prepare for the Swiggy ML interview?

Confirm the round list with your recruiter early; our outline comes from very few reports. Revise statistics and classical ML as seriously as LLMs; the breadth round is where one candidate was stopped. Drill SQL window functions and pandas groupby, rank and transform until they are automatic.

Are there free Swiggy interview questions?

Yes. LeetML has 12 questions reported from Swiggy interviews, and 3 of them are free to solve in the browser; they are listed by stage above.

Related interview guides

Official sources