Tiktok AI Research Scientist interview typically runs 4 rounds: HR screen, technical rounds, and final manager interview. It usually takes about 1-2 weeks and is notably research-heavy, with live coding and deep follow-up questions.
$166K
Avg. Base Comp
$297K
Avg. Total Comp
4-5
Typical Rounds
2-4 weeks
Process Length
We’ve seen TikTok’s AI Research Scientist interviews reward candidates who can defend their work at a level deeper than the abstract. Multiple candidates reported that interviewers kept pressing past the headline of a project into the study design, implementation details, and the reasoning behind specific choices. That pattern shows up again in the ML-heavy rounds, where the discussion moved quickly from fundamentals into architectures for multimodal language models and even diffusion-network specifics. The signal here is clear: surface-level familiarity is not enough; the team seems to care a lot about whether you can explain why a method works, where it breaks, and what tradeoffs you made.
Another recurring theme is how applied the technical bar feels. Our candidates report being asked to write pseudo-code for a transformer, implement a cross-attention block, and reason through graph traversal choices live, alongside questions about GPU bottlenecks and evaluation metrics. That mix suggests TikTok is looking for researchers who can move comfortably between theory and code, not just publish or prototype in isolation. We also noticed one unusual but important pattern: one candidate was unexpectedly interviewed in Mandarin with no warning, which implies language flexibility can matter in ways that aren’t always spelled out upfront.
What makes this process distinctive is the emphasis on composure under interruption and follow-up pressure. Interviewers repeatedly challenged answers in real time, and the strongest candidates were the ones who could stay precise while being pushed on details. In our view, TikTok is screening for people who can think like researchers, code like practitioners, and communicate clearly when the conversation gets very specific very fast.
Synthetized from 2 candidates reports by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the Tiktok process.
What caught me off guard most was that the whole interview was conducted in Mandarin with no warning at all. The interviewer started speaking Chinese immediately and expected me to answer in Chinese too, so I spent part of the round mentally switching gears before I could even get into the substance of the questions. After that, it was mostly a research-heavy conversation about my past work and study design, with the interviewer probing whether I really understood the details rather than just the headline results.
The process I went through was pretty consistent with that research-first style. In the first technical round, I spent about 30 minutes walking through past projects, and the interviewer kept interrupting with follow-up questions to test depth of understanding. The second half of that round was a LeetCode-style coding problem, specifically a dynamic programming question. Another technical round mixed ML problem-solving with coding and asked about architectures for multimodal language models. I also got a very detailed set of questions on diffusion networks, including things like naming two activations used in a diffusion network and explaining how they differ. In the coding portion of that round, I was asked to implement a cross-attention block, which was more of a practical coding test than a pure algorithms question.
There was also an HR screen at the start, and the final manager interview was more high-level, focused on research directions and the kinds of problems the team cares about. Overall, the interviews felt less like a standard machine learning loop and more like a mix of research defense, applied ML discussion, and coding under time pressure. I didn’t get an offer, so my main takeaway is that for this role you need to be ready to go very deep on your own research, explain design choices clearly, and code core ML components like attention blocks from scratch.
Prep tip from this candidate
Be ready to defend your past research in detail, especially study design and project-specific decisions, and practice implementing a cross-attention block by hand. It also helps to review diffusion-network internals and multimodal language model architectures, since those came up as deep technical follow-ups.
Share your own interview experience to unlock all reports, or subscribe for full access.
Sourced from candidate reports and verified by our team.
Topics based on recent interview experiences.
Featured question at Tiktok
Given two sorted lists, write a function to merge them into one sorted list.
| Question | |
|---|---|
| P-value to a Layman | |
| Hurdles In Data Projects | |
| Flatten N-Dimensional Array to 1D Array | |
| Basic Regex | |
| Bias vs. Variance Tradeoff | |
| Overfit Avoidance | |
| Unsafe Content ML Design | |
| Target Value Search | |
| Data Preparation for Imbalanced Data | |
| f(x,y) in Interval | |
| Data Cleaning Experiences | |
| Why Do You Want to Work With Us | |
| LRU Cache 1 | |
| Marketing Dollar Efficiency | |
| TikTok FYP Algorithm | |
| 2nd Highest Salary | |
| Experiment Validity | |
| Button AB Test | |
| Scrambled Tickets | |
| Bank Fraud Model | |
| Network Experiment Design | |
| Friendship Timeline | |
| Permutation Palindrome | |
| Swipe Precision | |
| Nearest Common Ancestor | |
| Using R Squared | |
| Testing Price Increase | |
| Good Grades and Favorite Colors | |
| Swiping App Design |
Synthesized from candidate reports. Individual experiences may vary.
An initial HR conversation starts the process. This is typically a quick screening to confirm background, motivation, and basic fit before moving into the technical loop.
The first technical round focuses heavily on past research and project experience. Interviewers probe deeply into your study design, implementation choices, and whether you truly understand the details behind your work, sometimes including a coding problem such as dynamic programming.
This round mixes machine learning fundamentals with hands-on coding. Candidates may be asked about classification, evaluation metrics, transformer architecture, GPU bottlenecks, or to write pseudocode for a transformer or solve a LeetCode-style problem.
Another technical round goes deeper into applied ML problem-solving and model internals. Topics mentioned include multimodal language model architectures, diffusion networks, and implementing core components like a cross-attention block from scratch.
The final manager conversation is higher level and centers on research direction and the kinds of problems the team works on. It is less about coding and more about how your interests and experience align with the team’s research priorities.