Software Development Lifecycle Flashcards
7 cards from real CAIC practice questions. Tap to flip, then mark Knew It or Still Learning — missed cards come back until you master them.
Read the first 7 Software Development Lifecycle flashcards as text
Which type of technical debt is UNIQUE to AI systems compared to traditional software?
Answer: Dependency on stale training data
AI systems accumulate data debt when training datasets become stale, causing model performance to degrade as the real world changes.
During sprint retrospective, a team notes that model retraining takes too long. The BEST solution within an Agile SDLC is:
Answer: Add automated retraining pipelines as backlog items
Creating backlog items for pipeline automation is the Agile approach to addressing process inefficiencies iteratively.
A feature store is BEST described as:
Answer: A centralized system for managing and sharing ML features across teams
A feature store centralizes the computation, storage, and serving of ML features, ensuring consistency between training and inference.
In an AI project, a 'shadow mode' deployment is used to:
Answer: Run a new model in parallel with the old one without affecting users
Shadow mode runs the new model alongside the production model, capturing its outputs for evaluation without impacting user experience.
Which gate is MOST appropriate to include in an AI CI/CD pipeline to prevent data quality regressions?
Answer: Automated data validation and schema checks
Automated data validation checks enforce schema, null rates, and distribution bounds so pipeline failures catch data quality issues immediately.
When managing an AI project with regulatory compliance requirements, which SDLC artifact is MOST important for an audit?
Answer: Model card documenting training data, performance, and limitations
A model card provides structured documentation regulators need: training data provenance, evaluation results, intended use, and known limitations.
In the context of AI SDLC, 'concept drift' requires which operational response?
Answer: Monitoring model performance metrics and triggering retraining
Concept drift means the statistical relationship between inputs and outputs has changed, requiring performance monitoring and model retraining.