ML System Design Templates

Use these templates to structure design discussions.

Template A: Ranking / Recommendations

• Goal and metrics (CTR, watch time, satisfaction)
• Candidate generation vs ranking
• Features and freshness
• Training pipeline and evaluation
• Serving latency and fallbacks
• Monitoring: drift, bias, feedback loops

Template B: Fraud / Abuse

• High recall vs high precision tradeoff
• Label acquisition and adversarial behavior
• Real-time scoring vs batch
• Human-in-the-loop
• Monitoring for concept drift and false positives

Template C: Forecasting

• horizon and granularity
• missing data and seasonality
• evaluation beyond MSE (business cost)
• retraining schedule