Meta System Design Interview Questions

The API layer must support real-time, bidirectional communication. Use WebSockets or a long-lived HTTP/2 connection to keep latency low. The API should handle message delivery guarantees (at least once), message ordering, and typing indicators. For scalability, shard user conversations and use a publish-subscribe pattern to fan out messages to recipients. Include presence updates and delivery/read receipts, with all APIs gated by authentication and rate limiting.

A notification system must be event-driven, scalable, and user-personalized. When an event occurs (new comment, tag, friend request), an event producer sends it to a notification service via a message bus. The service applies business rules (priority, deduplication, user preferences) and queues the message for delivery via push, email, or in-app. Multi-device sync ensures that dismissing a notification on one device removes it everywhere.

Recommendations, whether for friends, videos, or groups, are generated using hybrid algorithms: collaborative filtering for similarity, content-based filtering for relevance, and ranking models optimized via A/B testing. Data is processed in a feature store, with real-time updates from user actions. The system must support context-aware recommendations, factoring in time, location, and active trends, while enforcing diversity to avoid filter bubbles.

Ad analytics must ingest billions of events daily and make them queryable within seconds. Events (impressions, clicks, conversions) flow into Kafka, are processed with Flink/Spark Streaming for aggregation, and land in a low-latency OLAP store like Druid or Pinot. Partition by advertiser and campaign to speed queries. Retain raw logs for audits and backfills, while exposing aggregated metrics to dashboards and APIs.

Live streaming at scale uses CDN edge delivery with adaptive bitrate streaming (HLS/DASH). The video is captured, transcoded into multiple bitrates, chunked, and distributed. A control channel handles chat, reactions, and viewer counts in real time. The system must minimize end-to-end latency while ensuring stability during audience spikes. For viral broadcasts, the CDN should dynamically replicate streams closer to demand hotspots.

Live comments must handle thousands of messages per second. Use a pub-sub architecture where comments are published to a topic per live stream, and subscribed clients get updates instantly. To reduce bandwidth, batch messages or apply rate limits per user. Moderation filters run inline to block spam or abusive content before delivery. Clients may use sliding windows to show only recent comments for readability.

Moderation combines automated filtering with human review. The pipeline starts with AI models detecting policy violations (hate speech, nudity, misinformation). Detected content is flagged and either auto-removed or queued for review. Use tiered queues for priority cases (e.g., viral posts). Maintain an audit trail for transparency. The system must support multiple languages, cultural contexts, and evolving policy rules.