AI Project Life Cycle, MLOps, and DataOps CAIPM Practice Questions (Page 7)

Questions 31–35

1. 31application · medium

   A retail company has deployed a demand-forecasting model that retrains nightly. The data science team manually updates the training script and then runs a separate process to rebuild the model artifact. Recently, a change to the training script caused the model to fail in production because the artifact was built from an older version of the code. Which practice should the company adopt to prevent this class of failure?

   • AHave the operations team manually verify the artifact before each production deployment
   • BUse a CI/CD pipeline that builds the model artifact from the same commit that passed validation
   • CStore all model artifacts in a shared network drive for easy access
   • DIncrease the frequency of manual retraining to reduce the window for errors
   Reveal Solution Discussion Select an answer first
2. 32application · medium

   A financial services firm runs a daily batch pipeline that ingests transaction data, cleans it, and feeds it to a fraud-detection model. The data engineering team frequently receives complaints from the ML team that the data is stale or incomplete, but the data engineers only discover issues when the model's performance drops. Which DataOps practice should the firm implement to detect data quality problems earlier?

   • AHave the ML team manually inspect the data before each training run
   • BIncrease the frequency of model retraining to compensate for data issues
   • CSchedule a weekly meeting between data engineers and ML engineers
   • DAdd automated data quality checks with alerts at each pipeline stage
   Reveal Solution Discussion Select an answer first
4. 33foundation · easy

   Which best practice is recommended to address the challenge of data quality issues in DataOps pipelines?

   • ASkip data validation to speed up pipeline execution
   • BImplement automated data validation and monitoring at each stage of the pipeline
   • CManually inspect data after each pipeline run
   • DUse only historical data to avoid new data issues
   Reveal Solution Discussion Select an answer first
5. 34application · medium

   A bank wants to build a model to approve loan applications. The data science team has collected historical data and built a prototype that shows promising accuracy. Before deploying the model, the bank must ensure that the model does not discriminate against protected groups. In which phase of the AI project life cycle should this fairness evaluation be formally conducted?

   • ADeployment
   • BProblem definition
   • CModel validation and testing
   • DData collection
   Reveal Solution Discussion Select an answer first
6. 35expert · hard

   A large enterprise is adopting MLOps across multiple teams. The data science teams use different programming languages and frameworks. The operations team requires centralized governance and auditability. The program manager must choose an approach that balances team flexibility with governance. Which approach is BEST?

   • AUse a manual approval process for all model deployments.
   • BMandate a single programming language and framework for all teams.
   • CImplement a centralized MLOps platform with standardized CI/CD and model registry.
   • DAllow each team to manage its own MLOps tooling independently.
   Reveal Solution Discussion Select an answer first