AI0-001 AI Infrastructure and Technologies Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Apache Spark in batch mode

  Why it's wrong here

  Spark batch mode processes bounded datasets on a schedule, so clickstream events arriving continuously are only transformed after each job triggers, not per event. Spark Structured Streaming would fit; batch mode suits periodic reprocessing of landed historical data.

• ✗

  Snowflake

  Why it's wrong here

  Snowflake ingests streaming via Snowpipe but lands data in micro-partitions for SQL analytics; it cannot execute continuous per-event feature computation with sub-second latency. It suits batch or micro-batch warehousing, where near-real-time aggregates over staged data are acceptable.

• ✓

  Apache Kafka

  Why this is correct

  Kafka is a distributed publish-subscribe log that ingests continuous event streams with low latency and durable ordering, satisfying the real-time feature engineering requirement. Batch stores such as S3 or HDFS cannot process clickstream events as they arrive, so they fail the streaming constraint.

• ✗

  Apache Airflow

  Why it's wrong here

  Airflow orchestrates scheduled tasks; it triggers jobs at intervals rather than processing each event as it arrives, so feature freshness is bounded by the DAG schedule. It is the right tool for coordinating batch pipelines and dependencies, not continuous stream computation.