5.3 Explain AI Workflow Design and Monitoring Principles AI-TECHNICAL-PRACTITIONER Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   Which stage in an AI workflow is primarily responsible for transforming raw data into a format suitable for model training?

   • AModel training
   • BPreprocessing
   • CData ingestion
   • DDeployment
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A financial services firm is building an AI workflow that includes data ingestion from multiple sources, feature engineering, model training, and deployment to a production API. The compliance team requires that every step be logged and that the workflow be able to rerun from any failed stage without reprocessing earlier stages. Which orchestration feature is most important to implement?

   • AA retry policy that automatically restarts the entire workflow on any failure.
   • BParallel execution of all stages to reduce overall runtime.
   • CState management that tracks the completion status of each stage and allows resuming from the point of failure.
   • DA scheduling mechanism that runs the workflow at a fixed time each day.
   Reveal Solution Discussion Select an answer first
4. 23expert · hard

   A large enterprise is building an AI workflow for real-time fraud detection. The workflow has multiple stages: data ingestion from streaming sources, feature engineering, model inference, and alert generation. The compliance team requires that every alert be logged with the exact model version and input features used. The operations team wants to minimize latency between data ingestion and alert generation. The data science team wants to experiment with different feature sets without affecting the production workflow. Which architecture best meets all requirements?

   • AA batch processing system that runs every hour to generate alerts, with logging done after the fact.
   • BA modular pipeline with separate stages for ingestion, feature engineering, inference, and alerting, orchestrated by a workflow tool that logs each stage's inputs and outputs.
   • CA single serverless function that handles all stages, with logging to a central log file.
   • DA single streaming pipeline that processes data and generates alerts, with model version and features logged at each step.
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   A government agency is deploying an AI system to screen job applications. The agency must ensure that the system does not discriminate against protected groups and that its decisions can be explained. What should the agency do?

   • ARely on the model's overall accuracy as proof of fairness.
   • BConduct a fairness audit and implement explainability tools to understand model decisions.
   • CUse a proprietary model with high accuracy and keep its logic confidential.
   • DDelegate all decisions to the model without human oversight.
   Reveal Solution Discussion Select an answer first
6. 25foundation · easy

   In the design of an AI workflow, which stage typically comes immediately after model training?

   • APreprocessing
   • BEvaluation
   • CData ingestion
   • DDeployment
   Reveal Solution Discussion Select an answer first