InterviewCompany
Search-quality thinking, scalable systems, data reasoning, and ML fundamentals in a Google-style practice set.
Use prefix indexing, ranking, caching, and latency-aware serving.
A strong answer covers trie or inverted-index style retrieval, query logs, popularity ranking, personalization boundaries, cache layers, abuse controls, and freshness for trending queries.
Use offline metrics, human judgment, and online experiments together.
Offline metrics catch obvious regressions, human raters judge relevance, and A/B tests measure user behavior such as reformulation rate, clicks, and long-click satisfaction.
Split a large job into independent map tasks, then aggregate results in reduce tasks.
The important interview angle is fault tolerance and data locality: workers process chunks near storage, intermediate keys are shuffled, and reducers combine values for each key.
Check metrics by dependency, region, release, cache hit rate, and traffic shape.
Start with dashboards and recent changes. Break latency into network, queueing, application, database, and downstream calls. Then isolate whether the issue is global or localized.