Design a News Feed with APIs

Q: Design a News Feed with APIs

This question evaluates system design, scalable API design, and machine-learning ranking competencies for personalized content delivery, focusing on ingestion, candidate generation, ranking, and operational guarantees.

Q: How do I approach Machine Learning interview questions?

Machine Learning questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master machine learning interviews.

Q: What difficulty level is this interview question?

This is a hard difficulty Machine Learning question, commonly asked during Onsite rounds at Yelp.

Q: What role is this question designed for?

This question is commonly asked for Data Scientist candidates at Yelp during technical interviews.

Question

Personalized News Feed System Design (Push + Pull)

Context

You are designing a large-scale personalized news feed for a consumer application. The feed must support both push (server-initiated delivery/notifications) and pull (client-initiated fetch) consumption patterns, with strong ranking, safety, and operational guarantees.

Scale and Latency Targets

Users: 100M monthly active users (MAU)
Publishers: 1M
Ingestion: 200k writes/sec
Reads: 2M reads/sec
Latency: p99 < 200 ms for feed reads

Functional Requirements

Social graph and controls:
- Follow/unfollow
- Mute and block (user-level and publisher-level)
Feed quality:
- Deduplication (exact and near-duplicate)
- Diversity (content type/source/topic)
- Freshness/recency constraints
Push channel safeguards:
- Daily notification caps per user
Content lifecycle:
- Content retractions (publisher-initiated takedown)

API Design Deliverables

Design external APIs with request/response schemas and idempotency for:

publish
subscribe (follow)/unsubscribe
get_feed (pagination + consistent cursors)
ack_consume
retract
feedback logging

For each, provide:

Endpoint path and method
Request and response JSON schemas
Idempotency strategy

Architecture Choices and Justification

Choose fan-out-on-write vs. fan-out-on-read (or hybrid). Justify with the given scale and SLOs.
Describe storage layers: hot cache/timeline store vs. cold storage.
Describe event streaming topology.
Ranking pipeline: candidate generation, features, filtering, scoring, re-ranking, and online inference.

Reliability, Safety, and Operations

Abuse/spam controls
Multi-region replication and read locality
Backfill and replay strategy
GDPR deletion and data lifecycle
Rate limiting and quotas
SLOs, observability, on-call signals
Failure modes and graceful degradation (including partial outages)

Experimentation

How to run online experiments on ranking while ensuring user-level traffic consistency across sessions and pagination
Safe rollout plans and guardrails