Where to place feature transformations: client, serving API, or upstream service?
Tests separation of concerns in ML systems. Client causes duplication and skew; serving API couples compute to requests; dedicated service adds a network hop but centralizes logic. Red flag: ignoring training-serving skew.
Tests architectural tradeoffs between latency, consistency, and reuse in production ML systems. A strong answer compares all three placements. Client-side logic risks training-serving skew and duplication across languages. Colocating transforms in the serving API minimizes hops but bloats the critical inference path. A dedicated upstream service centralizes logic and reuse yet adds RPC latency and operational overhead. Red flag: proposing client-side transforms without discussing skew or insisting every transform belongs in one tier.
Read the original → developers.google.com
- #mlops
- #feature-engineering
- #system-design
- #production-ml
- #architecture
Get five bites like this every day.
Tezvyn delivers a daily feed of 60-second tech bites with quizzes to lock in what you learn.