Does TikTok use Kafka/Flink for real-time recommendation?
Introduction
The question “Does TikTok use Kafka/Flink for real-time recommendation?” sounds very reasonable, as real-time recommendation systems often need event streaming and stream processing.
But reasonable doesn’t mean correct.
Based on the public sources I checked, TikTok has official articles explaining the For You feed at the product/recommendation signal level, but I haven’t seen an official source explicitly state that TikTok uses Kafka or Flink in its recommendation pipeline.
So this post doesn’t assert that TikTok uses Kafka/Flink. This post simply separates: what is known, what is just a common pattern, and what shouldn’t be written as fact.
1. What do TikTok sources say?
TikTok says the For You feed is recommended by ranking videos based on multiple factors like user interactions, video information, and device/account settings. TikTok also has support documents saying recommender systems use preferences informed by interactions like following accounts or liking posts.
Thus, we know TikTok has a recommendation system that uses user behavior signals.
But these sources do not specify:
- whether Kafka is used;
- whether Flink is used;
- what the event streaming architecture looks like;
- what the feature store or online inference pipeline specifically entails.
2. What are Kafka and Flink usually used for?
Apache Kafka is an event streaming platform. It’s usually used to ingest, store, and process streams of events.
Apache Flink is a stream processing framework, used for processing data streams with state, window, event time, and real-time analytics/processing problems.
In a general real-time recommendation system, Kafka/Flink might appear in parts like:
User event
→ Event stream
→ Stream processing
→ Feature update
→ Online inference
→ Ranking resultBut this is a general pattern, not a definitive statement about TikTok.
3. How to write correctly
Do not write:
TikTok uses Kafka/Flink for real-time recommendation.Should write:
TikTok publicly states its For You feed is based on interaction signals and ranking. For real-time recommendation systems in general, event streaming and stream processing are two common technical blocks; Kafka and Flink are popular examples for these two technology groups. However, there are not enough public sources to assert that TikTok uses Kafka/Flink in its recommendation pipeline.Writing like this sounds less “cool”, but it is more accurate.
4. Lessons for small AI products
For a small CRM, you don’t need Kafka/Flink initially. If there are only a few thousand events/day, a simple queue or background job is enough.
When should you think about Kafka/Flink?
- large event volume;
- multiple services consuming events;
- need for replay/backfill;
- need for real-time analytics;
- need for stateful stream processing;
- batch pipelines are too slow for the product.
Example of a small CRM:
lead_created
→ background job enrich lead
→ AI scoring
→ update dashboardNo Kafka needed yet.
A larger CRM:
lead_created
message_sent
quote_opened
deal_updated
→ event stream
→ multiple consumers
→ real-time scoring
→ analytics dashboardAt this point, event streaming starts to make more sense.
5. Conclusion
A good technical blog post doesn’t need to pretend to know TikTok’s internal architecture. Just state the facts: what signals TikTok publishes, what kind of problems Kafka/Flink solve, and when a small system really needs them.
If you don’t know, say you don’t know. That’s also a part of engineering.
References
- TikTok Newsroom — How TikTok recommends videos #ForYou: https://newsroom.tiktok.com/en-us/how-tiktok-recommends-videos-for-you
- TikTok Support — How TikTok recommends content: https://support.tiktok.com/en/using-tiktok/exploring-videos/how-tiktok-recommends-content
- Apache Kafka — Introduction: https://kafka.apache.org/intro
- Apache Flink — What is Apache Flink?: https://flink.apache.org/what-is-flink/
New posts, shipping stories, and nerdy links straight to your inbox.
2x per month, pure signal, zero fluff.