E-News 2.0 — High-Scale Media Ecosystem

1. Problem

Traditional broadcast systems serve identical content to all users. Live TV traffic spikes (e.g. election results) crash article API servers. Personalisation is compute-heavy. A monolithic architecture cannot handle all three concerns simultaneously.

2. Domain-Driven Decomposition

• Feed Service (Golang) — stateless, read-heavy, aggressive Redis caching
• Streaming Service — HLS manifests, CDN offloading, short TTL
• Personalization Engine (Python) — async, multi-factor ranking algorithm
• Breaking News Layer — WebSocket overlays via Kafka event streaming

3. Deterministic Ranking (No Black-Box AI)

def compute_relevance_score(article, user_pref):
    cat_score   = user_pref["weights"].get(article["category"], 0)   # 40%
    bias_score  = 1.0 - abs(article["bias_val"] - user_pref["bias_val"])  # 30%
    hours_old   = (NOW - article["published_at"]).hours
    time_score  = 1.0 / (1.0 + 0.1 * hours_old)                     # 30%
    return 0.40 * cat_score + 0.30 * bias_score + 0.30 * time_score

4. Circuit Breaker Pattern

routes:
  - path: /feed/v1/*
    circuit_breaker:
      max_failures: 5
      timeout: 200ms
      fallback: generic-trending-feed   # graceful degradation
  - path: /stream/live/*
    caching:
      ttl: 5s   # short TTL for live manifests

5. Results

• 100K+ concurrent connections in production
• Fault isolation: each service fails independently
• Transparent, explainable ranking (no opaque ML scoring)