TikTok interview process is actually quite friendly. There is no pressure to answer purely algorithmic questions. It is more about asking in-depth questions about the projects you have done. From the time I submitted TikTok DS Intern in March to the time I got the offer in May, the whole process took more than two months, and I passed both rounds of VO smoothly. This article sorts out the entire interview process and the questions asked, providing a reference for those preparing for the TikTok DS internship later.
Tiktok data science intern basic information
- Post：Summer 2026 Data Science Intern
- Timeline: Official website delivery on March 14 → Appointment by email on March 16 → One appointment on March 21 → Second appointment in mid-April
- Interview format: All in English, video interview
- Interviewer: All are DS or Analytics Engineers with Chinese background and 3-5 years of working experience.
One side: project digging + SQL
After introducing yourself, start asking questions about the project
The interviewer was a Chinese girl, and she mainly asked in-depth questions about my two internships.
Internship 1: Dashboard + Text Classification
Asked me how to choose metrics, I said I chose DAU, Engagement Rate, and Conversion Rate, which correspond to the three dimensions of business scale, stickiness, and revenue. She pressed why she chose these three, and I explained what business problem each corresponded to.
For the text classification part, she asked how to determine accuracy. I talked about using Precision / Recall / F1-Score, and then she continued to ask what it depends on whether to choose Precision or Recall. I said it depends on the business cost - if the cost of accidentally killing users is high, it depends on Precision, if the cost of missing illegal content is high, it depends on Recall. She nodded.
Internship 2: ETL + data processing
The focus was on what the most technical job is. I talked about dealing with data deduplication and skewing.
In the original data, a user may have multiple logs a day, and counting distinct logs directly is very slow. My plan is to group by user_id and take the first one, and then use Spark broadcast to optimize the small table join.
She asked what if the data was larger, and I said we could use HyperLogLog to approximate the count. When asked how to determine the granularity of a partitioned table, I said partitioning by date because most queries are date ranges.
SQL questions
Calculate DAU and average online time. I first use CTE to calculate the two metrics separately, and then join them.
She asked how the duration is calculated if a user logs in multiple times a day. I said that if you want to calculate the real online time, you need to merge the overlapping time periods, but if you just simply average the current writing method, it will be fine. She said, "Okay, I think clearly."
The interviewer said that there is still a Python part later but there is not enough time. One side ends.
Side 2: Business Analysis + Case
Resume questions
When asked how to do data preprocessing, I said processing missing values, outliers, data normalization, etc. Asked how to understand Bias-Variance Tradeoff, I explained that high bias means under-fitting and high variance means over-fitting, which should be balanced through regularization and cross-validation.
Hypothesis Case
The examiner asked: If your product is going to be launched online, how will you study its effect?
I talked about the experimental design framework - defining goals, formulating hypotheses, A/B testing, measurement indicators, risk control, and launch strategies. For example, for this text classification model, I will A/B test 50% of the traffic using the new model to measure the review speed, error rate, and labor cost, and run it for 2 weeks.
She asked how to determine that two weeks is enough, and I said it depends on the historical data sample size and confidence requirements. If it fails, rollback and analyze the reason.
SQL remove duplicate questions
Output the total number of users who registered and posted on that day. Please note that one user may post multiple posts.
Sql
SELECT COUNT(DISTINCT user_id) AS user_count
FROM users u
JOIN posts p ON u.user_id = p.user_id
WHERE DATE(u.created_at) = DATE(p.created_at);She specifically emphasized that "many people will miss DISTINCT".
Business Case
Based on the above metric (users who registered and posted on the day), how to improve the engagement of creators?
I analyzed this metric, which reflects the initial activation rate of new users, and looked for problems—why there was no posting after registration, and why there was no interaction after posting. Then propose a plan - optimizing novice guidance, improving recommendation algorithms, and analyzing the characteristics of highly active groups for key operations. To measure effectiveness, look at the time from registration to first post and subsequent retention rate.
The examiner was very satisfied and said, "You not only looked at numbers, but also thought of business problems and solutions."
TikTok DS interview core suggestions
Your resume should tell a story: Every project must prepare "why we did this" and "what happened in the end"
SQL details matter: You will be asked about small things like DISTINCT, deduplication, and time processing.
Case analysis framework: From data problems → business problems → solutions → measuring effects, we need to have a clear system
English communication: Full interview in English, no need to be particularly fluent but clear logic
About my landing experience
After successfully passing the first level, I realized that the second level will test my business analysis and case capabilities more, which is where I am relatively weak. I was a little anxious at the time and didn’t know how to prepare systematically. Later, I asked the ProgramHelp team to provide VO assistance. Their team all have many years of DS experience in top foreign companies and are very familiar with interviews with big companies. The most important thing is that they have helped many students get ashore, including some of my classmates, and they have a good reputation.
Resume optimization, behavioral problem coaching, and VO assistance include:ProgramHelp · Dachang interview coaching
