Design a real-time home feed ranker
Company: Pinterest
Role: Machine Learning Engineer
Category: ML System Design
Difficulty: hard
Interview Round: Onsite
## Scenario
Design a **real-time home feed** (e.g., social or content platform) that is **responsive to user engagement**.
Users open the app and see a ranked list of posts/items. As they engage (click/like/comment/dwell), the feed should adapt quickly.
## Task
Design the end-to-end system including:
1. **Candidate generation** and **ranking**.
2. Incorporating **real-time engagement signals** (both the viewer’s actions and global engagement).
3. Storage, services, and streaming/infra components needed to meet latency and scale.
4. Model training, feature pipelines (offline/online), evaluation, and experimentation.
## Requirements (state assumptions if needed)
- **Latency:** low p95 end-to-end feed response.
- **Freshness:** newly published items can appear quickly.
- **Personalization:** based on user history + session context.
- **Reliability:** graceful degradation if real-time components fail.
- **Integrity:** prevent spam/manipulation; handle feedback loops.
Quick Answer: This question evaluates an engineer's ability to design scalable, low-latency real-time recommendation and ranking systems that integrate personalization, streaming engagement signals, candidate generation and ranking, feature pipelines, model training, and operational concerns such as reliability and integrity against manipulation.