TikTok System Design Interview Questions
Almost every TikTok design question reduces to three systems. Those are video upload and delivery, the For You feed, and the event pipeline that records what users do.
If you can draw those three and defend the trade-offs, you can answer most versions of the question. The rest is detail.
TikTok has publicly said it passed one billion monthly users. So every answer has to survive a read-heavy, global load. Say that out loud early.
The Questions That Get Asked
| Question | What it really tests | Hardest part |
|---|---|---|
| Design TikTok | End to end video platform | Scope control in 45 minutes |
| Design the For You feed | Ranking and serving at scale | Freshness against cost |
| Design video upload and playback | Storage, transcoding, delivery | Start time for playback |
| Design the like and view counter | High write volume | Approximate counts |
| Design live streaming with chat | Real-time delivery | Fan-out to many viewers |
| Design the comment system | Write path and ordering | Hot videos with heavy traffic |
Start by Cutting the Scope
The question is too large to answer fully. Narrowing it is part of the score.
Say which features you will build. Upload, feed, and playback is a good set. Push search, messaging, and ads out of scope, out loud.
Then state your numbers as assumptions. For example, 500 million daily users, 50 million uploads per day, and a read to write ratio near 1000 to 1. Nobody expects the real numbers. They expect you to show the math you build on.
Last, name the two properties that matter most. For TikTok those are low start time for a video and high availability. Full consistency on counts can wait.
The Upload and Delivery Path
A creator uploads a file to an upload service. That service writes the raw file to object storage and adds a job to a queue.
Workers pick up the job and transcode the video. Transcoding means producing several versions at different sizes and bit rates. This lets a weak network get a small file.
Finished files are pushed to a content delivery network, which is a set of servers near users. Playback then reads from the nearest copy, not from your origin.
Metadata goes in a separate store. Video id, author, caption, tags, and status belong in a database you can query, not in the file store.
Use a state field on the video record. It moves from uploaded, to processing, to ready. The client polls or receives a push when it turns ready.
The For You Feed
Split ranking into two stages. This is the part interviewers wait for.
Stage one is candidate generation. Pull a few thousand videos the user might like, using recent activity, followed accounts, tags, and trending lists. This stage is cheap and wide.
Stage two is ranking. Score those candidates with a model that uses watch time, replays, likes, and skips. This stage is expensive and narrow.
Then apply rules on top. Remove videos already seen. Mix in fresh content so new creators get views. Enforce safety filters.
Precompute part of the feed for active users and cache it. Compute the rest at request time. That mix keeps cost down while the feed still reacts within a session.
The Event Pipeline
Every view, like, skip, and share is an event. These events are what make the feed feel personal.
Send them to a log system such as Kafka. Several consumers then read the same stream.
One consumer updates counters. One updates the user profile used by ranking. One feeds offline model training. One feeds fraud and safety checks.
Counters do not need to be exact in real time. Update them in batches, and accept a short delay. Say this trade-off out loud, because it is the cheap answer to a very expensive problem.
Trade-Offs You Should Name
Storage cost against playback quality. More transcoded versions means faster starts and a larger bill.
Freshness against cost. Ranking every request is better and far more expensive than serving a cached feed.
Consistency against speed. View counts can lag by seconds. A payment could not.
Global reach against complexity. Regional storage and delivery cut latency and add data rules you must handle.
How to Prepare
- Learn the building blocks first. Caching, sharding, queues, and content delivery networks appear in every version of this question. Grokking System Design Fundamentals covers them.
- Practice the full question with a timer. Give yourself 45 minutes and draw the three systems. Grokking the System Design Interview works through video platforms in this format.
- Read the TikTok-specific versions. See what TikTok system design interview questions to expect and what the TikTok system design interview is like.
- Know the difference between rounds. Object design and system design are graded differently. See low-level design against system design questions.
- Compare with other feed products. Top system design questions for Meta reuse the same feed and fan-out ideas.
- Prepare the behavioral round too. See what is asked in a TikTok behavioral interview.
GET YOUR FREE
Coding Questions Catalog
$99
$197
$72
