Company interview guide
Yandex Machine Learning Interview Guide (2026)
Classic algorithm problems written without running the code, plus rapid-fire ML fundamentals and a domain deep dive for senior roles.
Yandex's machine learning interview loop has 5 reported stages (Recruiter call, Algorithms & coding, ML fundamentals & metrics, Specialization & ML system design, Final meetings with teams); typical length: 2-3 weeks (company estimate).
Roles covered: ML Engineer and Deep Learning Engineer. The question bank holds 8 questions reported from Yandex interviews, 1 of them free to solve in your browser.
Restructured 2024. In July 2024 Yandex N.V. sold its Russian businesses, which still operate as Yandex; the international parent became Nebius Group.
- Interview process length
- 2-3 weeks (company estimate)
- Stages
- 5
- Reports linked to bank questions
- 8
- Bank report dates
- Mar 2023–Jul 2025
- Questions in bank
- 8 (1 free)
- Last reviewed
- October 2026
How the Yandex interview loop works
A recruiter call, then three or four technical sections of about an hour each, held on one day or spread over several, with feedback a few days after each. The company's ML hiring page names the sections: ML and programming (ML concepts and metrics plus algorithm problems), coding (usually two algorithm problems in an online editor), an ML specialization deep dive (NLP, CV, speech, recommender systems or classic ML) for experienced candidates, and ML system design for those with architecture experience. Junior and intern loops in candidate reports are often a single interview that mixes one coding problem with ML theory. The loop ends with one or more meetings with team leads, after which you pick a team and receive an offer. Most of our reports date from 2022 and 2023, so treat the exact mix as a guide rather than a rule.
The Yandex interview process, stage by stage
01Recruiter call
Short video call with a recruiter; length not specified
The recruiter goes over your experience and interests, explains the open roles and lays out which technical sections you will get. The set of sections depends on your seniority, so this is where you learn whether a specialization or system design round is included.
What it tests
- Fit between your background and the open ML teams
- Which domain (NLP, CV, speech, recommender systems, classic ML) suits you
- Logistics: level, location, timeline
How to prepare
- Ask exactly which sections you will have and which programming language you may use.
- Name the domain you want the specialization section in, since it is chosen to match your background.
Only one candidate report mentions the recruiter round explicitly; the company's ML and analyst hiring pages both open the process with this call.
02Algorithms & coding
About 60 min in an online editor, usually two algorithm and data-structure problems; often you cannot run the code
The most consistently reported part of the loop. Problems are mostly easy to medium (two pointers, strings, intervals, arrays), though one senior candidate got edit distance. In junior and intern loops the problem often shares a single hour with ML theory questions; senior loops tend to have a separate section, and one 2025 report adds a hard SQL task. Interviewers check that the code is correct without help from a compiler.
What it tests
- Choosing an efficient algorithm and stating its complexity
- Writing correct code without running it
- Finding your own bugs by tracing through examples
- Two pointers, string and interval manipulation
How to prepare
- Practise easy and medium problems in a plain editor, then dry-run them on examples before saying you are done.
- Talk through the approach and complexity before writing code; the company's guidance asks for this.
- Write a simple working solution first rather than an ambitious one you may not finish.
- Keep dynamic programming classics like edit distance warm for senior loops.
Algorithm problems in 14 of 16 ML-role candidate reports (5 since 2024), mostly easy to medium; the company's ML hiring page describes a coding section with two problems.
03ML fundamentals & metrics
Part of the ML and programming section, often 30 min within a one-hour interview: oral questions, no coding
Fast, detailed questions on classical ML. Interviewers like to take one model you choose, often logistic regression, decision trees or gradient boosting, and press on its loss, optimization and failure modes, then move to evaluation metrics, overfitting and the bias-variance trade-off. Candidates who had not reviewed theory say it showed quickly.
What it tests
- Loss functions: logistic regression, cross-entropy, tree splitting criteria
- How gradient boosting builds its ensemble and what each tree contributes
- Classification metrics, ROC-AUC and how to plot it from a trained model
- Overfitting, regularization and bias versus variance across model families
- Convexity: whether gradient descent can get stuck on logistic regression
How to prepare
- Pick the model you would want to discuss and be able to derive its loss and gradient from scratch.
- Prepare a crisp explanation of gradient boosting, including what removing one tree does to predictions.
- Be ready for follow-up questions that probe edge cases rather than definitions.
Free reported questions for this stage
- Logistic Regression: Gradient, Hessian, and ConvexityStatistics & InferenceMediumSolve
ML theory questions in 11 of 16 ML-role candidate reports, but only 2 since 2024; the company's ML hiring page lists an ML and programming section on ML concepts and offline metrics.
04Specialization & ML system design
About 60 min each for experienced candidates: a domain deep dive built on a practical task, and/or an ML system design discussion
The company says senior candidates get a deep dive in their own domain and, with architecture experience, an ML system design round. Reported examples are a computer vision section that included fixing a deep learning model's training code, a marketplace matching and recommendation design, and an uplift-modeling pipeline for choosing which clients get a push notification, covered from data to A/B test.
What it tests
- Depth in one domain: NLP, CV, speech, recommender systems or classic ML
- Debugging model and training code
- Framing a business problem as an ML task with the right target and metrics
- End-to-end pipeline: data, features, model, deployment, online evaluation
How to prepare
- For classic ML, rehearse a ranking or uplift problem end to end, including the A/B test.
- For recommender systems, cover candidate generation, ranking, metrics and diversity.
- Review core deep learning building blocks such as convolution design choices if you claim CV.
Three ML-role candidate reports describe a specialization or ML system design round; the company's ML hiring page describes both sections and their domain topics.
05Final meetings with teams
One or more conversations with team leads and sometimes team members
Teams present their projects and discuss your experience to judge mutual fit, and you choose the team you want. The company says these meetings occasionally add technical questions when earlier sections left gaps, with notice beforehand; one candidate got a simple ML task here.
What it tests
- Depth and ownership in your recent projects
- Fit with the team's product and problems
- The questions you ask about the work
How to prepare
- Prepare two project stories with the task, your decisions, the metrics and the result.
- Use the meeting to compare teams; the offer follows your choice.
Two candidate reports describe final talks with team leads, mostly about experience; the company's ML hiring page describes these team meetings before the offer.
Questions and themes reported most often
- Talk about your previous experience or a recent project (4 reports, final meetings with teams)
- How gradient boosting works, and what removing one estimator does (3 reports, ml fundamentals & metrics)
- Classification metrics and how to build a ROC curve from a trained model (3 reports, ml fundamentals & metrics)
- Overfitting and the bias-variance trade-off across classical models (3 reports, ml fundamentals & metrics)
- Logistic regression: its loss and whether gradient descent can get stuck (2 reports, ml fundamentals & metrics)
- Decision trees, entropy and the tree loss function (2 reports, ml fundamentals & metrics)
- Minimum distance between elements of two arrays (1 reports, algorithms & coding)
- Valid palindrome after deleting at most one character (1 reports, algorithms & coding)
What Yandex asks about
- Algorithms & coding40%
- Classical ML theory & metrics30%
- Deep learning & specialization12%
- ML system design10%
- Experience & team fit8%
Yandex interview prep checklist
- Solved a set of easy and medium two-pointer, string and interval problems without running the code
- Can write edit distance with dynamic programming from memory
- Derived the logistic regression loss and gradient and explained why it is convex
- Explained gradient boosting step by step and what each tree corrects
- Can list classification metrics and sketch how a ROC curve is built from scores
- Walked through one ML product pipeline from target definition to A/B test
- Prepared two project stories with metrics and results
Tips
- Treat algorithms as a core round: even ML candidates were rejected for not finishing a medium problem.
- Practise solving problems in an editor without running code, then trace your own tests by hand.
- Be able to derive the loss and gradient of logistic regression and explain tree splits and boosting in detail.
- Know classification metrics cold, including how to compute ROC-AUC from model scores.
- Expect a terse style in some interviews with little small talk; keep explaining your reasoning anyway.
- If you are senior, prepare one domain deep enough for a practical deep dive and one end-to-end ML design.
- The company's own practice site and its hiring pages list topics and reading for each ML domain; use them.
- Prepare project stories with concrete numbers for the recruiter call and the team meetings.
See the rest of the Yandex loop
The later stages, the questions reported for them and how to prepare are in Pro.
Unlock with ProPractice the Yandex loop
Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.
Yandex interview FAQ
How many interview rounds does Yandex have for ML roles?
Candidates report 5 stages: Recruiter call, Algorithms & coding, ML fundamentals & metrics, Specialization & ML system design and Final meetings with teams. A recruiter call, then three or four technical sections of about an hour each, held on one day or spread over several, with feedback a few days after each. The company's ML hiring page names the sections: ML and programming (ML concepts and metrics plus algorithm problems), coding (usually two algorithm problems in an online editor), an ML specialization deep dive (NLP, CV, speech, recommender systems or classic ML) for experienced candidates, and ML system design for those with architecture experience. Junior and intern loops in candidate reports are often a single interview that mixes one coding problem with ML theory. The loop ends with one or more meetings with team leads, after which you pick a team and receive an offer. Most of our reports date from 2022 and 2023, so treat the exact mix as a guide rather than a rule.
How long does the Yandex ML interview process take?
Reported timelines run 2-3 weeks (company estimate).
What does Yandex ask in ML interviews?
By reported question mix: Algorithms & coding 40%; Classical ML theory & metrics 30%; Deep learning & specialization 12%; ML system design 10%; Experience & team fit 8%. The most repeated themes: Talk about your previous experience or a recent project; How gradient boosting works, and what removing one estimator does; Classification metrics and how to build a ROC curve from a trained model.
How should I prepare for the Yandex ML interview?
Treat algorithms as a core round: even ML candidates were rejected for not finishing a medium problem. Practise solving problems in an editor without running code, then trace your own tests by hand. Be able to derive the loss and gradient of logistic regression and explain tree splits and boosting in detail.
Are there free Yandex interview questions?
Yes. LeetML has 8 questions reported from Yandex interviews, and 1 of them are free to solve in the browser; they are listed by stage above.
Related interview guides
- GoogleClassic algorithm rounds plus ML design, ML domain depth and Googleyness, judged by a hiring committee.
- UberGrid and graph coding, ML from scratch and marketplace system design built around ETA, pricing and Uber Eats.
- Booking.comCase-driven ML and data science interviews on real travel-marketplace problems, with a coding track for ML engineers.
- SpotifyPractical Python and SQL, music-flavoured modeling cases and recommendation design, with a real values round.
- AmazonLeadership Principle questions in every round, standard algorithm coding, and ML breadth, depth and design rounds that differ by track.
- MicrosoftProject deep dives in most rounds, quick ML and LLM fundamentals, standard coding, from-scratch ML code and GenAI design, in a loop shaped by the hiring team.