TikTok Interview Questions — Page 4

TikTok·Machine Learning Engineer·Onsite - System Design / Architecture

Apr 2026

System design interview at TikTok for an MLE role, focused entirely on a two-part scenario: deploying a multimodal video captioning model under tight GPU/VRAM constraints, then building a retrieval and watermarking pipeline on top of it at billion-video scale. Dense and pretty unforgiving if you haven't thought about inference cost before.

Design an end-to-end system to deploy a multimodal video captioning model under hard compute and GPU memory constraints. How do you reduce input cost, fit the model in memory, structure the serving path, and monitor the whole thing?
For the captioning system, what levers do you have to reduce GPU memory usage, and what does each one cost you?
Design a retrieval pipeline that lets a brand advertiser find relevant videos from a corpus of hundreds of millions to billions of videos using a text or creative query, then watermark the matched videos at scale.
How do you keep the ANN index fresh as new videos are continuously ingested without doing a full rebuild every time, and what staleness would an advertiser actually see?
If a 4-bit quantized captioning model produces noticeably worse captions for a subset of videos, how would you detect those cases and route them to a higher-quality path?
Pick one ML fundamentals topic (dropout, normalization variants, or RL in LLM post-training) and explain how it connects to deploying or fine-tuning the captioning model.

“This is the kind of question where you can spiral if you start with the model instead of the input.”

View Post

TikTok·Product Manager·Onsite - Product Sense / Strategy

Apr 2026

Interviewed for a PM role at TikTok and got asked to compare the three big short/long-form video platforms. Pretty standard product sense territory but TikTok asking you to size up their own product against competitors is a bit of a weird vibe.

What are the key differences between TikTok, Instagram, and YouTube as platforms?

“Felt a little strange answering this to someone who works at TikTok.”

View Post

TikTok·Software Engineer·Technical Phone Screen

Apr 2026

TikTok software engineer interview with a classic LRU cache problem. Nothing too wild but the O(1) constraint is where people trip up if they haven't seen it before.

Design and implement an LRU cache with a fixed capacity, supporting get and put operations both in O(1) time.

“The problem itself is straightforward if you've seen it, but I fumbled explaining why a plain hashmap isn't enough.”

View Post

TikTok·Software Engineer·Technical Phone Screen

Apr 2026

TikTok data engineer technical screen, two-part grid/island problem. Part A was pretty standard BFS stuff but the no-in-place-modification constraint tripped me up for a minute. Part B was the real test and I don't think I fully nailed the shape encoding.

Given a binary grid where 1s are land and 0s are water, count the number of islands using BFS. You cannot modify the grid in place to track visited cells.
Follow-up: given the same grid, return the count of distinct island shapes, where two islands are the same shape if one can be translated to overlap the other (no rotations or reflections).

“The BFS part was fine, I've done island counting before.”

View Post

TikTok·Data Scientist·Onsite - Behavioral / Leadership

Apr 2026

Behavioral portion of a TikTok Data Scientist interview, the kind where they give you a scenario and you pick the most and least effective response from a list. Pretty formulaic but it tripped me up more than I expected.

In a scenario where a teammate keeps missing deadlines, walk through why you'd pick a particular response as the most effective and another as the least effective.

“I kept second-guessing myself because two of the options both seemed reasonable.”

View Post

TikTok·Data Scientist·Technical Phone Screen

Apr 2026

TikTok data scientist interview with a product analytics case built around an Apple phone launch. The question felt straightforward on the surface but there's a lot of room to either impress or ramble depending on how structured you are.

Apple just launched a new phone. What are the most important metrics you'd include in a TikTok launch performance dashboard for them, and why does each one matter?

“I started rattling off reach and impressions and then kind of stalled.”

View Post

TikTok·Product Manager·Onsite - Product Sense / Strategy

Apr 2026

TikTok PM interview with a product design question aimed at a non-traditional user base. Short and focused, basically one question that sounds deceptively simple but has a lot of moving parts if you actually think it through.

How would you redesign TikTok specifically for elderly users?

“My first instinct was to just say 'bigger text, simpler UI' and I almost stopped there which would've been embarrassing.”

View Post

TikTok·Data Scientist·Technical Phone Screen

Apr 2026

TikTok Data Scientist interview with a scheduling algorithm problem dressed up as a game scenario. The core was a real algorithmic challenge about job dispatching across parallel machines, so don't let the 'Production Factory' framing fool you into thinking it's a product case.

You have jobs arriving with different processing times that need to be assigned to one of two parallel machines in real time. How do you design an algorithm that minimizes average flow time across all jobs?

“I knew greedy was the right instinct but fumbled explaining why shortest-processing-time works for flow time specifically versus makespan.”

View Post

TikTok·Software Engineer·Onsite - System Design / Architecture

Apr 2026

TikTok system design round where they basically want you to own a project top to bottom, no hand-holding. The question is deceptively broad but the real pressure comes from the follow-ups on whichever component they decide to dig into.

Pick a recent project and walk through its architecture end-to-end. Start with a high-level diagram, then go deep on the design, the trade-offs you made, and how you'd evolve it. Expect follow-ups on any single component.
For the system you described, walk through the key APIs and data flows. Which paths are synchronous versus asynchronous, and what protocols did you use?
Justify the major design decisions you made and explain what alternatives you rejected. How do constraints like latency, throughput, and cost factor into each choice?
How does your system scale? Walk through your capacity planning, including QPS, bandwidth, and compute. Quantify the key constraints.
How does your system handle partial failures? Talk through replication, circuit breakers, backpressure, idempotency, retry policies, and dead-letter queues.
Describe your observability setup across both synchronous and async boundaries. What SLOs did you track and how did you manage error budgets?
Walk through the security model for your system. How did you handle authentication, authorization, secrets management, and PII?
Describe a real incident you experienced with this system. What was the symptom, the root cause, how did you detect it, how did you mitigate it, and what was the permanent fix?
What concrete results did your work on this system produce? Share numbers around latency, reliability, quality, or cost.
Looking back, what would you redesign in this system and why?

“This is the whole interview, not just a warmup.”

View Post

TikTok·Data Scientist·Technical Phone Screen

Apr 2026

TikTok Data Scientist interview that was basically a coding/algorithms problem dressed up in an e-commerce scenario. One meaty question, fairly well-defined constraints, and the whole thing hinged on whether you could think through preprocessing and binary search under pressure.

Given N products and M customers, design and implement an algorithm where each customer finds the maximum number of distinct products they can buy within their budget. Ties in count are broken by smallest total spend, then lexicographically smallest list of product IDs. Preprocessing should run in O(N log N) and each query in O(log N). Justify correctness.

“The greedy insight is straightforward once you see it: sort products by price ascending (breaking ties by product ID), build a prefix sum array, then for each customer budget do a binary search to find how many you can afford from the front of that sorted list.”

View Post