The TikTok Coding Interview: Questions and Format

TikTok is a product built inside ByteDance, and its coding loop reflects a company that ships the For You feed, TikTok LIVE, and creator tools at enormous scale. It hires engineers who can move at that pace. For a backend or general software role, plan for a hard timed online assessment out front, one or two technical phone screens, then two coding rounds inside the sequential onsite. Each round often carries two problems. Add it up and you solve a lot across the loop. That volume, more than any single trick, is the tax.

The problems themselves are mostly plain. Live rounds are frequently drawn straight from the standard algorithm canon, and candidates describe them as high-frequency mediums with nothing exotic in them. The pressure comes from the count, the clock, and the bar for finished code. It does not come from decoding a disguised product scenario. That shapes how you prepare.

The TikTok angle

The two gates have opposite difficulty profiles, and most candidates misjudge which one to fear. The online assessment is where TikTok is genuinely hard. You get multiple problems on a tight timer, hidden test cases with partial scoring, and a real tail of contest-flavored work. That tail looks like a bitmask over a tiny column count, a prefix-XOR identity, a hard dynamic-programming state, or a circular simulation that punishes a naive loop. The live rounds are the calmer half, recurring as two mediums with clean, well-known shapes. Prepare the assessment like it is the whole interview, because it is the round that quietly ends most loops before a human ever sees you.

How the coding rounds run

The loop is sequential. You pass each round to reach the next, with roughly a week between them and feedback in a few business days, so failing any single round ends the process there. Interviews run on Lark or Zoom, and interviewers are often based in Singapore or China. Round counts flex a lot by team and level, so read the shape below as the modal picture rather than a fixed script. TikTok publishes little about its hiring, so confirm team-specific details with your recruiter.

• Online assessment: CodeSignal, when present. Around four problems in 70 to 90 minutes, camera on and screen shared, no AI assistance. Point-weighted so the later problems dominate the score, and hidden cases give partial credit. Migrated from HackerRank. Experienced or referred candidates often skip it and start at a phone screen.
• Live coding: Two rounds, ~2 problems each. One hour each, typically two mediums drawn from high-frequency shapes. The interviewer wants full running code that passes the corner cases, plus a complexity read. Working and tested beats clever and broken.
• Paired discussion: 15 to 30 min on your work. A coding slot often opens with a short resume or past-project dive before the problem, especially for senior and machine-learning tracks. The hour is rarely pure coding end to end.
• Team variance: The biggest unknown. Same title, different loop. Trust and Safety and some infra teams lean on clean, practical coding over pure algorithms. Machine-learning and data-engineering tracks front-load a domain deep-dive and ask fewer algorithm problems. Confirm your track with the recruiter.

The TikTok angle

Running, tested code is the explicit bar. Interviewers ask for a full implementation that passes thorough test cases, plus the time and space complexity. Then they hand you a case you did not consider, to see whether you handle it. A solution that compiles and clears the happy path is the start of the round, not the end of it. Walk your own edge cases before you are prompted, keep the code executing as you go, and state the complexity out loud. On a loop where you clear many problems under time, the habit of finishing each one cleanly is what separates a pass from a near miss.

Where the difficulty lives

The online assessment is the round to over-prepare. Its problems are point-weighted, so the last two are worth far more than the first two. The hardest of them reach past the usual medium tier into work that rewards a sharper idea than brute force. Partial passes are common on this tail: an obvious quadratic solution clears a fraction of the hidden cases and times out on the rest. The product framing is only surface, a problem set among collectibles or a viral challenge. Read past it on the first pass and the algorithm underneath is a standard number-theory, prefix-XOR, or dynamic-programming problem.

One myth worth killing early: that TikTok barely asks dynamic programming. It does. The recurring set includes two Premium questions, plus a genuinely hard subsequence problem on the assessment. Graphs and topological sort appear too, and so do interval problems. Treat every family as fair game rather than betting the loop on a narrow slice.

What the rounds ask

Weighting your prep by how often a family comes up beats grinding a single tier. The distribution below is directional rather than exact. TikTok publishes no such count, so read it as a weighting, not a measurement. It also tilts harder and mathier inside the assessment than in the live rounds. Use it to decide where your hours go.

The TikTok angle

Because the problems are plain, you prepare for difficulty and volume. There is no product scenario to decode. Consumer-scale peers often bury the pattern inside a delivery or booking prompt you have to unpack first; TikTok mostly does not. Live rounds recur as two Premium questions, Clone Graph, and a Premium question, often close to verbatim. So do not spend time imagining a TikTok-specific version of each pattern or hunting for a secret company twist. Spend it building fluency and speed across the standard shapes, because you will meet several of them back to back and the room is timing how cleanly you close each one.

The follow-up that escalates

A pattern worth rehearsing on its own: a plain problem that grows a second act. You build an inventory checkout or a Premium question, get it working, and then the interviewer asks how it fails under many callers at once. Now you are reasoning about a race condition, reaching for a mutex, and then refining toward fine-grained per-item locking when a single global lock throttles everything. No new algorithm arrives; the same code has to survive concurrent access. This shows up strongly for mobile and backend candidates, and it rewards knowing your language's locking primitives cold.

The TikTok angle

Data-structure-design prompts are the other escalation, and they favor candidates who have built the structure rather than only used it. Phone screens reach for a custom hash map with an O(1) operation that sets every value at once. Or a Premium question extended with a time-to-live, so entries expire as well as evict. Or a typeahead trie that returns the top completions for a prefix. Each starts simple, then stacks a requirement that breaks the obvious first design. Practice these until you can implement the core from memory. The round grades how you extend it under pressure, not whether you recognize the name.

How to prepare for this

Build speed and coverage first. You face many problems across the assessment and two live rounds, so the goal is to solve any standard medium cleanly and fast, with edge cases and complexity handled without prompting. Mastering one exotic family matters less than that fluency. Drill the recurring shapes in a timer, write out the corner cases before you call a solution done, and say the complexity aloud so it becomes reflex.

Then harden the two things that trip candidates up here. Spend real time on the assessment's harder tail, so a bitmask over a small state space or a prefix-XOR reduction is a tool you reach for rather than rediscover under the clock. And rehearse the follow-up ladders: take a working a Premium question and make it thread-safe, then add expiry; take a trie and add ranked completions. If your track is machine-learning, data engineering, or Trust and Safety, weight your prep toward that track's loop, since the algorithm surface is lighter there and the domain depth is heavier.

What this section covers

Five pattern lessons follow, each taking one part of the map above and working its recurring core as runnable problems, with the follow-ups the room tends to add. Where a lesson rests on a foundational data-structures idea, the deep walkthrough lives in the free Foundations course so these pages stay on what is specifically TikTok. Start with arrays, strings, and hashing, the widest bucket and the warm-up most loops open on.

Common questions