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 metric is MOST appropriate for tracking the health of a deployed AI model over time?
Answer: Model performance drift relative to baseline
Tracking performance drift against a baseline detects degradation due to concept or data drift, triggering retraining when thresholds are breached.
A team is conducting a 'post-mortem' after an AI model caused a significant error in production. What SDLC artifact is MOST useful as input to this review?
Answer: Model monitoring logs and prediction audit trail
Monitoring logs and audit trails provide the factual record of model inputs, outputs, and decisions needed to diagnose root causes in a post-mortem.
When transitioning an AI project from development to production, which risk is MOST often underestimated?
Answer: Operational infrastructure and model serving latency requirements
Teams often underestimate the engineering work needed to serve models reliably at scale, including latency, availability, and integration requirements.
In Kanban applied to an AI team, Work In Progress (WIP) limits PRIMARILY help by:
Answer: Reducing bottlenecks and improving flow of tasks through the pipeline
WIP limits force the team to finish in-progress work before starting new items, surfacing bottlenecks and improving overall throughput.
Which approach BEST handles changing business requirements mid-project in an AI SDLC?
Answer: Use iterative sprints with regular backlog refinement to adapt incrementally
Iterative development with regular backlog refinement allows teams to incorporate changing requirements without derailing the entire project.
An AI consultant recommends adding an 'explainability layer' to a credit-scoring model. In the SDLC, this requirement should be captured in:
Answer: Functional and non-functional requirements during the requirements phase
Explainability is a non-functional requirement (regulatory and ethical) that must be captured in requirements to be designed, built, and tested properly.
Which testing type is SPECIFIC to AI systems and has no direct equivalent in traditional software testing?
Answer: Adversarial robustness testing
Adversarial robustness testing evaluates how an AI model behaves under intentionally crafted inputs designed to fool or destabilize it, with no traditional software parallel.