Company interview guide

TikTok Machine Learning Interview Guide (2026)

Hour-long mixed rounds: a resume grilling, ML fundamentals and an algorithm problem, repeated two to four times.

TikTok's machine learning interview loop has 9 reported stages, from hr screen to data scientist track: sql, stats & product case; typical length: 2-8 weeks.

Roles covered: ML Engineer, Research Scientist, Applied Scientist and Data Scientist. The question bank holds 57 questions reported from TikTok interviews, 9 of them free to solve in your browser.

Try a free question

US joint venture. In January 2026 TikTok's US business moved into a US joint venture owned mainly by US investors, with ByteDance keeping a minority stake. US roles may now be hired by the joint venture; reports from other regions are unaffected.

Interview process length
2-8 weeks
Stages
9
Reports linked to bank questions
83
Bank report dates
Feb 2022–Sep 2026
Questions in bank
57 (9 free)
Last reviewed
October 2026

How the TikTok interview loop works

An HR screen, an online assessment for some candidates (mostly interns and new grads), then two to four technical rounds of 45-60 minutes held one at a time; each must be passed before the next is booked. Most technical rounds follow the same pattern: resume deep dive, ML knowledge questions, then one or two algorithm problems, with some teams turning one or two rounds into ML system design. A hiring manager round usually comes last, sometimes followed by a director or cross-team leader and an HR or HRBP close. A hiring freeze or a move to another team can restart part of the loop. Research tracks spend more time on papers, and a few candidates gave a research presentation.

The TikTok interview process, stage by stage

01HR screen

15-30 min call with HR or a recruiter

A short call on your background, location, timing and why you want to join. Recruiters often reach out first, and some describe the later rounds or match you to a team before interviews start; ask, because team and round format vary widely. Several candidates say the recruiter could not tell them what the rounds would cover.

What it tests

  • Why TikTok, and how your work fits the team
  • Logistics: location, start date, level, compensation expectations
  • Occasionally one or two high-level ML questions

How to prepare

  • Ask which team you are interviewing for and whether rounds will include ML system design; that decides most of your prep.
  • Ask whether interviews may be held in Mandarin so it does not surprise you.
  • Ask for the number of rounds and the level; several candidates were left guessing on both.

An HR or recruiter call opens the loop in 34 candidate reports across ML and data scientist roles, mostly 2024-2026; the company's early careers page also lists recruiter phone pre-screening before interviews.

02Online assessment

About 1-2 hours on a proctored coding platform: 2-3 algorithm problems plus multiple-choice questions

Mainly for interns and new grads; referred and experienced candidates often skip it. Expect medium to hard dynamic programming or graph problems with hidden test cases, plus quick multiple-choice questions on ML fundamentals and sometimes SQL, Linux or distributed systems.

What it tests

  • Medium-hard dynamic programming and graph problems under time pressure
  • Handling hidden test cases and edge cases
  • ML fundamentals in multiple-choice form: metrics, feature engineering, model evaluation

How to prepare

  • Budget time per problem; candidates report one hard problem eating the whole session.
  • Test edge cases yourself before submitting, since debugging tools may be limited.

Described in 26 candidate reports, about half interns or new grads; referred candidates often skip it. The company's careers pages say technical roles may get an online coding assessment, by role and region.

03Technical rounds: resume and research deep dive

20-45 min of each 45-60 min technical round, one interviewer

Nearly every technical round starts here, and it often takes most of the hour. Interviewers walk your projects in order and ask for your exact contribution, the math and data behind each model, why you made each choice and what you would change; some try to reproduce details. Research candidates spend more of each round on papers, and a few were asked for a short research presentation.

What it tests

  • Your exact personal contribution on each project
  • The math, data and metrics behind your models
  • Why you chose each method over the alternatives, and what failed
  • How your past work maps to the team's domain

How to prepare

  • For each project prepare the problem, data, model, metric, results, limits and what you would redo.
  • Remove anything from your resume you cannot discuss at depth; the questions follow what you list.
  • Expect the same projects to come up in every round, and keep your answers consistent.

The most consistent element: about 130 ML and data scientist reports describe a resume or research deep dive inside technical rounds, often 30-45 minutes of an hour-long round.

04Technical rounds: ML fundamentals & ML coding

10-30 min of each technical round, spoken Q&A, sometimes writing code or pseudocode

Rapid-fire knowledge questions, often triggered by what is on your resume, with Transformer and attention questions asked most. Some rounds ask you to write attention, cross-entropy or logistic regression from scratch while answering follow-ups as you go.

What it tests

  • Transformer blocks, self-attention shapes, why scores are scaled by the key dimension, the role of the FFN
  • Normalization: batch norm vs layer norm, and which one LLMs use and why
  • Overfitting, dropout at train and inference time, class imbalance and resampling
  • Classic ML: logistic regression loss, cross-entropy, tree ensembles, precision and recall
  • Writing attention, cross-entropy or gradient descent from scratch

How to prepare

  • Be able to derive and explain every model and loss on your resume; questions follow your own projects.
  • Practise writing single-head and multi-head attention plus cross-entropy on a blank page without running code.
  • Prepare short answers on LLM fine-tuning and post-training (SFT, LoRA, PPO, DPO, GRPO) and on how LLMs can help recommendation; these have come up recently.

Free reported questions for this stage

About 120 ML-role reports describe ML, deep learning or LLM knowledge questions in technical rounds; 16 reports since 2024 add a short ML coding task such as attention, cross-entropy or logistic regression.

05Technical rounds: coding

15-30 min at the end of each 45-60 min technical round, shared editor; usually one medium algorithm problem, sometimes a hard one or two problems

Almost every technical round ends with an algorithm problem, and coding performance is often described as the deciding factor. Dynamic programming, backtracking, prefix sums, sliding windows, trees, graphs and stack designs dominate, often with a follow-up that tightens time or space.

What it tests

  • Dynamic programming: subsequences, reachability, non-adjacent sums, segmentations
  • Backtracking such as N-Queens, counted or enumerated
  • Prefix sums and sliding windows with an O(1)-space follow-up
  • Binary search, trees and BSTs, BFS or DFS on grids
  • Stack designs (min or max stack) and streaming follow-ups

How to prepare

  • Go straight for the optimal solution; reports say brute force alone does not pass.
  • Expect a variant twist on a familiar problem, such as returning the second-best answer or adding a non-negative constraint.
  • Time is short because the resume discussion runs long, so practise solving a medium in 15-20 minutes.
  • Read the problem carefully and confirm inputs; interviewers do not always explain clearly.

About 220 ML-role reports include one or two algorithm problems per technical round, mostly medium with some hard; several say every technical round followed the same resume, ML and coding pattern.

06ML system design

30-60 min, either a full round (often with a team lead) or half of a technical round

Open-ended designs tied to the hiring team's product: short-video feed recommendation, ranking model internals, content moderation, search suggestions or multimodal models. Questions can go deep into one component, such as what the query, key and value should be in a cross-attention ranker or how to choose retrieval depth per request.

What it tests

  • End-to-end recommender design: retrieval, ranking, features, cold start, feedback loops
  • Ranking model internals: user-history attention, offline metrics, retrieval depth
  • Harmful content and abuse detection with human review
  • Search autocomplete and query suggestion optimized for CTR
  • Serving large multimodal models under compute and memory limits

How to prepare

  • Prepare a For You style recommender blueprint and be ready to zoom into any stage.
  • Tie every component to a metric, and name both online and offline evaluation.
  • Study the hiring team's product area; designs and case questions are usually drawn from it.

Free reported questions for this stage

ML system design appears in 34 ML-role reports (31 since 2024), either as a full round or half of a technical round; most other loops never mention it, so it depends on the team.

07Hiring manager round

45-60 min with the hiring manager or team lead, sometimes followed by a director or cross-team leader

Usually the last technical conversation. Managers revisit your projects at a higher level, test fit with the team's current problems and, in several reports, add a coding problem, ML questions or a design case from their business. Some candidates describe sharp pushback on every choice that felt like a pressure test. Research managers discuss research direction and ideas for their area.

What it tests

  • Fit with the team's domain and current problems
  • Defending design decisions under pushback
  • Business cases from the team's product
  • Research direction and ideas (research tracks)

How to prepare

  • Read up on the team's product area; managers often bring a case from it.
  • Stay calm and concrete under challenge; explain trade-offs rather than conceding or arguing.
  • Have one or two questions about the team's current priorities ready.

A hiring manager or team lead round is described in 33 ML and research reports, usually as the last technical round; 6 mention an extra director or cross-team leader conversation.

08HR / culture round

20-60 min with HR, an HRBP or a culture-fit interviewer

A lighter close on motivation, teamwork and expectations before an offer discussion. Behavioral questions are less structured than at most large tech companies.

What it tests

  • Why TikTok and why this team
  • Collaboration and handling conflict across teams
  • Your biggest challenge and how you handled it

How to prepare

  • Have one or two short stories on conflict and a hard challenge ready.
  • Follow up with the recruiter if you hear nothing; slow responses after the final round are common.

A closing HR, HRBP or behavioral round appears in 22 reports across ML and data scientist roles; other loops end at the hiring manager round and go straight to an offer call.

09Data scientist track: SQL, stats & product case

Data scientist loops: recruiter call, sometimes an online assessment or short written product exercise, then 2-4 rounds of 45-60 min mixing resume deep dive, SQL, product cases and A/B testing

Data scientist interviews replace algorithm coding with SQL and replace system design with product cases. Common prompts ask how to investigate a sudden metric change, how to design and size an experiment, and how to measure impact when no randomized test was run.

What it tests

  • Medium SQL with joins and window functions across several tables
  • Metric diagnosis and product sense for the TikTok app
  • A/B test design: metrics, guardrails, sample size, duration
  • Basic statistics and ML concepts, plus causal inference without randomization

How to prepare

  • Practise explaining a metric drop or spike step by step before jumping to causes.
  • Expect cases drawn from the team's own area, such as ads, shop, content reports or retention.
  • Know how network effects between creators and viewers can break a standard A/B test.

Based on 25 data scientist reports from 2025-2026: SQL in about 20, product or metric cases in about 20 and A/B testing in about 12; one mentions a short written product take-home.

Questions and themes reported most often

  • Deep dive into your resume projects with detailed follow-ups (120 reports, technical rounds: resume and research deep dive)
  • Explain the Transformer and self-attention, including the scaling by key dimension (35 reports, technical rounds: ml fundamentals & ml coding)
  • Dynamic programming: LIS, frog jump, non-adjacent sum, decodings (20 reports, technical rounds: coding)
  • Design an end-to-end recommendation system for a feed or shop (18 reports, ml system design)
  • A/B test design, sample size and metric diagnosis (data scientist) (15 reports, data scientist track: sql, stats & product case)
  • Batch norm vs layer norm, dropout and how to fight overfitting (10 reports, technical rounds: ml fundamentals & ml coding)
  • Implement self-attention or multi-head attention from scratch (9 reports, technical rounds: ml fundamentals & ml coding)
  • Binary search with a follow-up that extends or optimizes it (7 reports, technical rounds: coding)

What TikTok asks about

  • General coding30%
  • Resume deep dive & behavioral25%
  • ML & LLM concepts20%
  • ML / recsys system design13%
  • ML coding6%
  • SQL & experimentation (DS)6%

TikTok interview prep checklist

  • Can walk through every resume project: problem, data, model, metric, results, limits, what you would change
  • Solved N-Queens, word break and subarray-sum variants at optimal complexity in under 20 minutes each
  • Wrote multi-head attention and cross-entropy from scratch without running code
  • Can explain batch norm vs layer norm, dropout at inference and why attention is scaled
  • Sketched a short-video recommender end to end and zoomed into the ranking model
  • Have a specific answer to why TikTok and this team

Tips

  • Treat the resume as the syllabus: much of each hour is spent on your projects, so rehearse each one down to the math and the data.
  • Drill medium dynamic programming, backtracking and binary search until you can reach the optimal solution in about 15 minutes.
  • Have crisp, two-minute answers on Transformers, attention scaling, normalization, dropout and class imbalance; these repeat across rounds.
  • Write attention, cross-entropy and a logistic regression update from scratch on paper at least once before the loop.
  • Prepare one recommender design you can extend to ranking internals, moderation or search depending on the team.
  • If you speak Mandarin, practise explaining your projects in it too; a minority of rounds switch languages without warning.
  • Expect overlapping questions across rounds and answer consistently each time.
  • Each round is eliminatory, so treat every one as the decider rather than saving effort for a final onsite.

See the rest of the TikTok loop

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

Unlock with Pro

Practice the TikTok loop

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

TikTok interview FAQ

How many interview rounds does TikTok have for ML roles?

Candidates report 9 stages: HR screen, Online assessment, Technical rounds: resume and research deep dive, Technical rounds: ML fundamentals & ML coding, Technical rounds: coding, ML system design, Hiring manager round, HR / culture round and Data scientist track: SQL, stats & product case. An HR screen, an online assessment for some candidates (mostly interns and new grads), then two to four technical rounds of 45-60 minutes held one at a time; each must be passed before the next is booked. Most technical rounds follow the same pattern: resume deep dive, ML knowledge questions, then one or two algorithm problems, with some teams turning one or two rounds into ML system design. A hiring manager round usually comes last, sometimes followed by a director or cross-team leader and an HR or HRBP close. A hiring freeze or a move to another team can restart part of the loop. Research tracks spend more time on papers, and a few candidates gave a research presentation.

How long does the TikTok ML interview process take?

Reported timelines run 2-8 weeks.

What does TikTok ask in ML interviews?

By reported question mix: General coding 30%; Resume deep dive & behavioral 25%; ML & LLM concepts 20%; ML / recsys system design 13%; ML coding 6%; SQL & experimentation (DS) 6%. The most repeated themes: Deep dive into your resume projects with detailed follow-ups; Explain the Transformer and self-attention, including the scaling by key dimension; Dynamic programming: LIS, frog jump, non-adjacent sum, decodings.

How should I prepare for the TikTok ML interview?

Treat the resume as the syllabus: much of each hour is spent on your projects, so rehearse each one down to the math and the data. Drill medium dynamic programming, backtracking and binary search until you can reach the optimal solution in about 15 minutes. Have crisp, two-minute answers on Transformers, attention scaling, normalization, dropout and class imbalance; these repeat across rounds.

Are there free TikTok interview questions?

Yes. LeetML has 57 questions reported from TikTok interviews, and 9 of them are free to solve in the browser; they are listed by stage above.

Related interview guides

Official sources