Software Development Lifecycle Flashcards

Which SDLC phase is MOST critical for identifying AI model bias before a system goes to production?

Answer: Testing and validation

Testing and validation is the phase where bias, fairness, and accuracy issues are systematically identified and corrected before release.

In an AI project using Scrum, what artifact BEST captures the evolving data requirements for model training?

Answer: Product backlog

The product backlog holds all prioritized work items, including data acquisition and preprocessing tasks, and is continuously refined.

A team discovers that their AI model performs well in development but poorly in production. This is BEST described as:

Answer: Train-serve skew

Train-serve skew occurs when the data distribution or feature engineering in production differs from the training environment.

Which practice BEST supports continuous integration in an AI development pipeline?

Answer: Automated testing of data pipelines and model performance

Automated testing of data pipelines and model metrics ensures that each code or data change is validated before merging.

When defining the 'Definition of Done' for an AI feature, which criterion is MOST relevant?

Answer: Model accuracy meets the agreed threshold on a held-out test set

A measurable performance threshold on unseen data provides an objective, agreed criterion for completion of an AI feature.

In the AI SDLC, 'data versioning' primarily serves to:

Answer: Enable reproducibility of experiments and audits

Data versioning allows teams to reproduce any past experiment by associating model versions with the exact dataset snapshot used.

A company wants to adopt MLOps. Which SDLC stage does MLOps MOST directly extend?

Answer: Deployment and operations

MLOps applies DevOps principles to the deployment and ongoing operations of ML models, automating retraining, monitoring, and delivery.