Bloomberg System Design: Global News Aggregation & Trending

You are asked to design the backend for a global news platform that ingests articles from publishers and the Newser API (which only returns one region per call), processes real-time user interactions, computes trending scores, and serves a low-latency, personalized news feed and search to users worldwide.

Key challenges you must address include aggregating across regions despite Newser's per-region call limit, respecting API rate limits and quotas, supporting publisher push ingestion, and calculating real-time trending with time-decay while keeping results consistent and fast.

High-level flow/stages you should cover:

• Regioned ingestion: parallel region workers, regional fetch scheduler, and a deduplication layer to merge duplicate articles.
• Stream processing: durable message bus (e.g., Kafka) and stream processors (Flink/Beam) to compute time-decayed trending scores and materialize views.
• Storage & indexing: a write-optimized store for raw articles, a search index (Elasticsearch/OpenSearch) for keyword queries, and a fast cache (Redis + CDN) for feeds.
• Serving layer: feed and search APIs, personalization filters (language, interest, region), and push updates (WebSockets/SSE) for real-time UI.

Skill signals interviewers look for: distributed systems and scaling patterns, stream processing and windowing, search indexing and real-time index updates, API design and rate-limit strategies, data modeling for time-decay metrics, caching and cache invalidation, and operational concerns (monitoring, retries, fault tolerance).

In your design, justify trade-offs for consistency vs. latency, strategies to reduce Newser API cost (batching, caching, region prioritization), and how you would test correctness and recover from failures.