Design a Trending Content System

Clarified what "trending" entities are (hashtags/topics/posts) and freshness/staleness tolerance before designing
Identified scale (high write throughput) as the core constraint driving approximate algorithms
Explicitly scoped out ML personalization/deep NLP as extensions, keeping focus on counting/ranking pipeline

Proposed an event-driven pipeline: ingestion → durable queue (Kafka) → stream processing → aggregated store → cache/serving layer
Separated the write-heavy aggregation path from the read-heavy low-latency serving path (cache with short TTL)
Designed API and schema reflecting region/category segmentation and windowed scores, not just raw counts

Discussed approximate counting techniques (Count-Min Sketch, Space-Saving/heavy-hitters) and why exact counts aren't feasible at scale
Addressed time decay / sliding windows so trends reflect recent velocity, not cumulative all-time volume
Covered spam/bot manipulation defense with a concrete signal (diversity of users/IPs, growth rate anomaly)
Addressed hot-key/hot-partition skew for viral entities and a mitigation (salting/pre-aggregation)
Articulated clear trade-offs (accuracy vs memory, freshness vs infra complexity, consistency vs simplicity) rather than asserting one "correct" answer

Managed time across phases, reaching deep dives without over-spending on requirements or basic CRUD
Used the interviewer's cues/questions to prioritize which deep dive to go deepest on
Summarized design concisely at the end with a clear opening-statement-style recap

Problem