TikTok Interview Questions — Page 7

TikTok·Data Scientist·Technical Phone Screen

May 2026

First round screen for a Data Scientist role at TikTok. Pretty quick, just a probability warm-up to see if you know your distributions.

If X follows a Uniform(0,1) distribution, what is the expected number of independent draws needed before you get a value greater than 0.8?

“The answer is 5, geometric distribution with p=0.2.”

View Post

TikTok·Software Engineer·Technical Phone Screen

May 2026

TikTok software engineering interview that went deep on LLM fine-tuning, specifically DPO. Two meaty questions, both required real ML background to answer well. Not the typical coding grind I was expecting.

Explain Direct Preference Optimization at a high level and how it compares to a PPO-based RLHF pipeline. What objective does DPO optimize, why does it skip a separate reward model, and what are the practical trade-offs?
How would you construct a training dataset for DPO fine-tuning? Walk through the format of a single training example, how you'd collect preferred versus dispreferred responses, how you'd handle noisy labels or ties, and what preprocessing or filtering you'd apply.

“This is where I felt most comfortable but also where I probably over-explained.”

View Post

TikTok·Data Scientist·Technical Phone Screen

May 2026

TikTok data science interview that went deep into statistical inference. One question, but it was a beast. The kind of thing where you think you understand Simpson's paradox until someone asks you to prove it formally and then derive conditions to prevent it.

If average daily TikTok usage increases separately within both male and female user groups, does the overall average necessarily increase too? Walk through a formal proof or counterexample, the conditions under which the overall average is guaranteed to move in the same direction, how shifts in group composition affect the result, and what diagnostic checks you'd run on real data.

“I knew Simpson's paradox existed but building a concrete numerical counterexample on the spot was rougher than expected.”

View Post

TikTok·Data Scientist·Technical Phone Screen

May 2026

TikTok data science interview that went deep into causal inference territory. The whole session was basically one long PSM question broken into six parts, and if you haven't touched Rosenbaum bounds recently you will feel it.

You have observational data where users self-select into a feature. Walk through how you'd implement propensity score matching to estimate the average treatment effect on the treated for 7-day retention, starting with your choice of propensity model and why.
Describe your matching strategy: 1-to-1 nearest neighbor with or without replacement, how you'd set the caliper, and how you'd tune these choices.
What balance diagnostics would you run after matching, and what do you do if balance is insufficient?
How do you detect and handle lack of overlap in the propensity score distributions between treated and control groups?
For variance estimation after PSM, when would you use Abadie-Imbens standard errors versus bootstrapping?
Explain Rosenbaum sensitivity analysis: what is the gamma parameter, how do you find the point where your conclusions flip, and how would you communicate that to a non-technical product manager?

“I went with logistic regression first and framed it as the default when you want interpretability and your covariates are well-behaved.”

View Post

TikTok·Data Scientist·Technical Phone Screen

May 2026

TikTok data scientist interview with a SQL-heavy question around ad conversion metrics. Pretty straightforward setup but the metric definition piece tripped me up more than I expected.

Given a user events table with ad clicks and page visits, write SQL to compute the conversion rate of ad clicks that are followed by at least one page visit from the same user on the same calendar day. Define your numerator and denominator explicitly.

“The SQL itself wasn't the hard part.”

View Post

TikTok·Software Engineer·Onsite - Behavioral / Leadership

May 2026

Behavioral round for a software engineer role at TikTok, basically one big question dressed up as several. They want the full project story, not just a highlight reel.

Walk me through a challenging project from start to finish: the goal, the data, the stakeholders, your specific role, the hardest problems you faced, the trade-offs you made, how you validated the solution, and what measurable results it produced.

“This is the kind of question where you think you're prepared and then realize halfway through your answer that you've been narrating a team accomplishment instead of your own.”

View Post

TikTok·Data Scientist·Technical Phone Screen

May 2026

TikTok data scientist interview that leaned heavily into the product-meets-risk space. The core scenario was a classic internal disagreement between product and risk teams, and they wanted to see how you'd think through the tradeoffs without just picking a side.

Product and risk teams are at odds over how aggressively to block suspicious transactions, since being too aggressive hurts real customers. How do you find the right balance, and what metrics would you monitor over time to know if it's working?

“This one took me a second to get into because it sounds like a soft stakeholder question until you realize they actually want you to get into the numbers.”

View Post

TikTok·Data Scientist·Onsite - Behavioral / Leadership

May 2026

TikTok Data Scientist interview that centered almost entirely on a behavioral scenario about reading a disengaged stakeholder and recovering the situation. Pretty unusual angle for a DS role and I wasn't totally sure what they were actually testing.

Tell me about a time you sensed a hiring manager or key stakeholder had already made up their mind against your proposal before the meeting started. What specific signals tipped you off, and what did you do in the first few minutes to test that without making things awkward?
Walk through the exact steps you took to salvage a meeting that was going nowhere. Did you reframe the goals, propose cutting the agenda short, or try to surface one real problem you could solve on the spot? What did you actually say?
How do you stay professional and hold your ground when a meeting is clearly a waste of time? Where's the line between pushing for real engagement and just cutting your losses?
What tangible outcome came from that situation? Did you get a follow-up, useful rejection feedback, or any future opportunity? How did you document it afterward?
If you were the one running that meeting and realized the candidate or presenter wasn't getting a fair shot, what would you do differently to avoid wasting their time while still leaving them with a decent impression of the company?

“This one is harder than it sounds because the signals are almost always ambiguous.”

View Post

TikTok·Software Engineer·Technical Phone Screen

May 2026

Two-part coding interview for a software engineer role at TikTok. Both problems were algorithmic and required better-than-brute-force solutions with complexity justifications. Not a vibe check at all, they wanted real analysis.

Given an integer array where each element represents a measurement for a given day, return an array where each position holds the number of days you'd have to wait until a strictly higher measurement appears. If no such day exists, use 0. Solve it in O(n) time and justify correctness and space complexity.
Given a binary tree with unique node values and an integer k, return all unordered pairs of distinct nodes whose distance equals k, where distance counts intermediate nodes on the path. Your solution must be faster than O(n^2). Describe your data structures, analyze complexity, and handle edge cases like k = 0 or an empty tree.

“Classic monotonic stack problem once you recognize it, but I didn't recognize it fast enough.”

View Post

TikTok·Data Scientist·Technical Phone Screen

May 2026

TikTok data scientist case interview, heavy on fraud detection and promo design. Two hours of nonstop scenario work built around a flash sale abuse case. Technically demanding and honestly pretty well constructed as far as case questions go.

An account places 80 orders in 10 minutes across multiple cards and devices during a flash sale. What hypotheses besides fraud could explain this behavior?
What specific data would you pull in the next 15 minutes to distinguish coordinated fraud from organic high-volume buying? Name the features and joins.
Design a real-time approve/hold/reject decisioning strategy using a cost matrix where false positives cost $30 in lost GMV, false negatives cost $120 in fraud loss, and manual review costs $1 per order.
Quantify the GMV and loss impact of holding 50% of suspicious traffic for 30 minutes versus rejecting it outright. State your assumptions explicitly.
Redesign the flash sale promotion to reduce abuse. Cover coupon structure, per-user caps, velocity limits, payment risk tiers, address trust scores, and bot mitigation. Define guardrail metrics with alert thresholds.
Outline an experiment and monitoring plan for the redesigned promotion, including holdout or geo-split design, success and stop criteria, and explicit rollback conditions.

“I jumped straight to fraud and had to walk it back.”

View Post