Company interview guide

Deloitte Machine Learning Interview Guide (2026)

Resume-driven technical rounds on ML, GenAI and Python, then case, manager and partner conversations that vary by member firm.

Deloitte's machine learning interview loop has 6 reported stages (Recruiter or HR screen, Online assessment (graduate, campus and some offices), Technical interviews: projects, ML, GenAI and coding, Take-home or on-site test (some teams), Case study or business scenario, Manager, director, partner or client round); typical length: 1-8 weeks.

Roles covered: Data Scientist, AI Engineer, Gen AI Engineer and ML Engineer. The question bank holds 24 questions reported from Deloitte interviews, 1 of them free to solve in your browser.

Try a free question

Interview process length
1-8 weeks
Stages
6
Reports linked to bank questions
26
Bank report dates
Aug 2021–Mar 2025
Questions in bank
24 (1 free)
Last reviewed
October 2026

How the Deloitte interview loop works

Most experienced-hire reports describe two to four conversations: a recruiter or HR screen, one or two technical interviews (often 45-60 minutes each) that start from your projects and move to ML, statistics or GenAI questions with a little Python or SQL, then a manager, director or partner round, sometimes with an HR wrap-up at the end instead of the start. Some offices add a take-home or an on-site written test, a few client-facing roles end with an interview by the client, and campus and graduate routes begin with online aptitude, video or coding assessments. The company's pages describe two to three interview rounds with behavioral, technical or case content depending on the role, and an offer usually within one to two weeks of the final interview. Timelines in reports range from a one-day walk-in drive to about two months.

The Deloitte interview process, stage by stage

01Recruiter or HR screen

15-30 min call with a recruiter, talent-acquisition specialist or third-party agency; occasionally the team lead calls first

A first conversation about your background, motivation, notice period and salary expectations. Several candidates say level and pay came up bluntly here, and a few were offered a lower level than they applied for. In some Indian campus and walk-in drives the HR conversation comes at the end instead.

What it tests

  • Why Deloitte and why this practice
  • Fit between your experience and the job description
  • Logistics: notice period, location, expected salary and level

How to prepare

  • Have a clear salary range and a short answer for why you are moving; both are asked early.
  • Ask how many rounds follow, whether any is a take-home, case or client interview, and who runs each.
  • Know which offering the role sits in (for example AI & Data) so you can tie your projects to it.

Recruiter, HR or screening call first in 20 DS, AI and ML candidate reports (17 since 2024). The company's hiring pages list an initial recruiter screen covering background, motivation and pay.

02Online assessment (graduate, campus and some offices)

Online aptitude and behavioral tests, recorded video answers, or a timed coding test of 2-4 Python problems; some offices add an AI / ML quiz

Graduate and campus routes usually start with an untimed behavioral and reasoning assessment, then a job simulation with ranked choices and recorded video answers, before a final stage. Experienced-hire reports from India, Egypt and Israel more often describe a coding test or a mixed exam with Python, AI and ML questions. Several candidates were rejected straight from the test with no interview.

What it tests

  • Numerical, verbal, logical and abstract reasoning
  • Situational judgement: ranking what you would do in a workplace scenario
  • Short Python problems on lists, strings and basic algorithms
  • Quick AI / ML questions, for example which model suits video object detection

How to prepare

  • Read data-interpretation questions slowly; candidates say the multi-source data questions are the hard part.
  • For video answers, prepare two-minute STAR stories on achievements, teamwork and a decision you made.
  • Practise easy to medium Python problems in a plain editor against the clock.

Online tests in 15 DS, AI and ML reports (11 since 2024): aptitude or video in graduate routes, coding or AI quizzes elsewhere. The company's early-careers page describes the online assessment and job simulation.

03Technical interviews: projects, ML, GenAI and coding

1-2 rounds of about 45-60 min by video or in person with one or two practitioners or the team manager; coding, if any, is short and live

The core of the loop. Interviewers usually open with your resume and dig into every project: why that model, what the data looked like, how it was evaluated and deployed. They then move to concept questions, classical ML and statistics for data-science roles, and RAG, agents, transformers and prompting for AI and GenAI roles, often framed as client scenarios. About a third of reports mention one or two easy Python problems or SQL queries; a few say there was no coding at all.

What it tests

  • Depth on your own projects: model choice, data cleaning, evaluation, deployment
  • Classical ML: overfitting, bias-variance, trees, Naive Bayes, imbalance and metrics
  • Statistics: hypothesis tests, p-values, Type I and II errors
  • GenAI: how RAG and agents work, chunking, embeddings, attention, LLM evaluation
  • Easy Python or SQL written live

How to prepare

  • Prepare every project on your resume to the level of why you chose each model and what you would change; candidates repeatedly say anything listed is fair game.
  • Give the textbook definition first, then the practical nuance; some interviewers expect standard answers.
  • For AI roles, be ready to sketch a RAG pipeline end to end and explain RAG versus an agent.
  • Practise explaining a model choice to a non-technical client in plain words.

Free reported questions for this stage

Technical rounds in 53 DS, AI and ML reports (43 since 2024), mostly one or two led by resume projects; GenAI topics in about 20, live Python or SQL in about 19. Hiring pages say technical rounds depend on role.

04Take-home or on-site test (some teams)

A take-home notebook or coding exercise, sometimes with slides, or a 2-4 hour written test in the office; one long take-home took the candidate several days

Not every loop has one, but reports since 2024 describe classification take-homes, a drug-discovery modelling task, notebook-plus-slides assignments and a client-set coding exercise. One office ran a long on-site test: an easy coding task, open questions on LLMs, RAG and vision, then fixing a broken PyTorch training script and tuning an LLM prompt. Expect to present or defend your work afterwards.

What it tests

  • Building and evaluating a model end to end on a realistic problem
  • Finding bugs in training code: wrong validation loader, missing eval mode, wrong loss averaging
  • Reasoning about why a vision model fails in the real world
  • Presenting results clearly to a mixed audience

How to prepare

  • Ask for the time budget and how the work will be judged before you start; one candidate spent a week on a task with no stated standard.
  • Keep a short written summary and a few slides ready; several formats end with a presentation.
  • In a debugging task, check data loaders, train / eval mode and how the loss is averaged first.

Take-home or on-site practical test in 10 DS and AI reports, all since 2024; formats differ by office and team. The company's interview page says selected roles include a practical assessment.

05Case study or business scenario

A business case with a manager or partner, often 30 min; some roles get several short cases in one sitting

Consulting cases appear in a minority of data and AI loops, more often in Europe, the Middle East and campus drives. Prompts are client-style: a data migration strategy, how to add AI to a small business, what is wrong with a dataset, or several quick business cases with a partner. Candidates who were rejected here say they had not prepared for case questions at all.

What it tests

  • Structuring a vague client problem and stating assumptions
  • Turning business goals into a data or AI solution
  • Spotting issues in a dataset or dashboard and drawing conclusions
  • Explaining recommendations to a non-technical client

How to prepare

  • Use the company's own five steps: clarify the issue, state assumptions, summarize findings, recommend, outline next steps.
  • Practise at least two business cases out loud even for a technical role.
  • Close with measurable impact and a next step, not only a model choice.

A case round in 8 DS and AI reports (4 since 2024); many more mention scenario questions inside technical rounds. The company's case interview page and hiring pages describe case interviews for many roles.

06Manager, director, partner or client round

30-60 min with the hiring manager, a director, a partner or a senior leader; client-facing roles may add an interview with the client; HR may close the loop

The last conversation is usually about fit, the role, teamwork and how you handle clients, with some technical follow-up in the managerial rounds of Indian GenAI loops. A few AI roles placed on a client project ended with a client interview, and every round can be eliminating. Questions on where you see yourself in five years, strengths and weaknesses and why you are leaving recur.

What it tests

  • Motivation, career plans and fit with the team
  • Handling clients, ambiguity and disagreement
  • Seniority signals: leading delivery, mentoring, scoping
  • Follow-up on your projects and on transformer or RAG basics in some loops

How to prepare

  • Prepare STAR stories; the company says it uses behavioural event interviewing focused on specific past examples.
  • Have a crisp answer for where you see yourself in five years and why consulting.
  • If a client interview is scheduled, read about the client's industry and frame your projects in its terms.

A manager, director, partner, client or behavioral round in 24 DS, AI and ML reports (18 since 2024); two AI roles ended with a client interview. The company's hiring pages describe a final senior leader or partner interview.

Questions and themes reported most often

  • Walk through a resume project in depth: model choice, data, evaluation, deployment (24 reports, technical interviews: projects, ml, genai and coding)
  • What RAG is, chunking, building a knowledge base and evaluating it (10 reports, technical interviews: projects, ml, genai and coding)
  • Easy live Python: lists, primes, even and odd numbers, Fibonacci, mutable arguments (10 reports, technical interviews: projects, ml, genai and coding)
  • Why Deloitte, five-year plan, strengths and weaknesses (8 reports, manager, director, partner or client round)
  • SQL queries alongside the ML questions (7 reports, technical interviews: projects, ml, genai and coding)
  • Overfitting, bias-variance and choosing a model you can explain to a client (5 reports, technical interviews: projects, ml, genai and coding)
  • Statistics basics: hypothesis tests, p-values, Type I and II errors (5 reports, technical interviews: projects, ml, genai and coding)
  • Long take-home on a client-style modelling problem, then present it (5 reports, take-home or on-site test (some teams))

What Deloitte asks about

  • Projects & behavioral27%
  • ML & statistics concepts24%
  • GenAI, LLMs & RAG22%
  • Python & SQL coding15%
  • Case & business scenarios12%

Deloitte interview prep checklist

  • Can defend every project on your resume: data, model choice, metrics, deployment and what you would change
  • Explained overfitting, bias-variance, imbalance handling and F1 without notes
  • Sketched a RAG pipeline with an evaluation plan and compared it with an agent design
  • Solved a few easy Python list and string problems and two SQL queries in a plain editor
  • Answered hypothesis-test and Type I / II error questions in plain words
  • Practised one business case using the five-step structure
  • Prepared STAR stories plus answers on why Deloitte and where you see yourself in five years

Tips

  • Treat your resume as the syllabus: every project and tool on it can be probed in depth.
  • Revise classical ML and statistics definitions until you can give the standard answer and one practical caveat for each.
  • For AI and GenAI roles, prepare RAG end to end (chunking, embeddings, retrieval, evaluation) and a clear RAG versus agent comparison.
  • Keep Python and SQL basics warm; the coding is usually easy, but it is often asked without warning.
  • Practise one consulting case out loud using clarify, assume, summarize, recommend, next steps.
  • Ask the recruiter early whether the loop includes a take-home, an on-site test, a partner round or a client interview.
  • Frame answers in client terms: business impact, how you would explain the model, and how it would be deployed.
  • Expect slow feedback in some offices and follow up politely after each round.

See the rest of the Deloitte loop

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

Unlock with Pro

Practice the Deloitte loop

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

Deloitte interview FAQ

How many interview rounds does Deloitte have for ML roles?

Candidates report 6 stages: Recruiter or HR screen, Online assessment (graduate, campus and some offices), Technical interviews: projects, ML, GenAI and coding, Take-home or on-site test (some teams), Case study or business scenario and Manager, director, partner or client round. Most experienced-hire reports describe two to four conversations: a recruiter or HR screen, one or two technical interviews (often 45-60 minutes each) that start from your projects and move to ML, statistics or GenAI questions with a little Python or SQL, then a manager, director or partner round, sometimes with an HR wrap-up at the end instead of the start. Some offices add a take-home or an on-site written test, a few client-facing roles end with an interview by the client, and campus and graduate routes begin with online aptitude, video or coding assessments. The company's pages describe two to three interview rounds with behavioral, technical or case content depending on the role, and an offer usually within one to two weeks of the final interview. Timelines in reports range from a one-day walk-in drive to about two months.

How long does the Deloitte ML interview process take?

Reported timelines run 1-8 weeks.

What does Deloitte ask in ML interviews?

By reported question mix: Projects & behavioral 27%; ML & statistics concepts 24%; GenAI, LLMs & RAG 22%; Python & SQL coding 15%; Case & business scenarios 12%. The most repeated themes: Walk through a resume project in depth: model choice, data, evaluation, deployment; What RAG is, chunking, building a knowledge base and evaluating it; Easy live Python: lists, primes, even and odd numbers, Fibonacci, mutable arguments.

How should I prepare for the Deloitte ML interview?

Treat your resume as the syllabus: every project and tool on it can be probed in depth. Revise classical ML and statistics definitions until you can give the standard answer and one practical caveat for each. For AI and GenAI roles, prepare RAG end to end (chunking, embeddings, retrieval, evaluation) and a clear RAG versus agent comparison.

Are there free Deloitte interview questions?

Yes. LeetML has 24 questions reported from Deloitte interviews, and 1 of them are free to solve in the browser; they are listed by stage above.

Related interview guides

Official sources