AIMentor — AI Expertise Learning Platform

13 questions · pass mark 80% · 20 minute limit. The timer auto-submits at zero.

Time remaining: 20:00

Question 1easy
What do CI and CD stand for / do?
Code Input / Code DeliveryContinuous Integration (checks on every change) and Continuous Delivery/Deployment (ship the passed artifact)Container Init / Container DeployCompute Instance / Cloud Disk
Question 2easy
What is the AI-specific addition to a standard CI/CD pipeline?
A GPU checkAn eval gate — the golden-set regression suite runs alongside unit tests, because passing tests don't prove model qualityA larger base imageDisabling tests
Question 3medium
Why are unit tests necessary but not sufficient for an AI app?
They run too slowlyThey verify the code runs correctly but not that the model's output quality is still goodThey can't be automatedThey replace the need for eval
Question 4medium
How does an eval gate block a bad merge in CI?
It emails the teamThe eval script exits non-zero when quality drops below threshold, failing the CI jobIt deletes the branchIt increases temperature
Question 5hard
What cautions apply to an eval gate in CI? (Select all that apply)
It calls the model, so it costs tokens and adds timeIt's non-deterministic, so too-tight thresholds can flap the buildPin the model version and use a curated golden set with threshold marginCommit the API key into the workflow file for convenience
Select all that apply.
Question 6medium
Why tag deployable images by git commit SHA rather than using 'latest'?
SHAs are shorterA SHA tag is immutable and traces to exact code, so any version can be pinned and redeployed — the basis of rollback'latest' is not allowed by registriesIt reduces image size
Question 7medium
Where do secrets (like API keys) come from in a CI pipeline?
Hardcoded in the workflow YAMLThe CI platform's secret store (e.g. ${{ secrets.OPENAI_API_KEY }}), never committedBaked into the base imagePrinted to the build log
Question 8hard
What is a blue-green deployment?
Deploying to two clouds simultaneouslyRunning two environments; deploy/test on the idle one, then switch all traffic at once — rollback = switch backColoring the logsA/B testing prompts
Question 9hard
What triggers a rollback during a canary deploy of an AI app?
The build finishingOnline metrics (quality signal, error rate, p95 latency, cost) on the canary dipping vs. the current versionA successful unit testThe image being large
Question 10medium
How does CI/CD relate to the LLMOps discipline from Track 5?
They are unrelatedCI/CD is where the LLMOps principles (gates, versioning, safe rollout, rollback) become automationCI/CD replaces evaluationLLMOps only applies to training
Question 11hard
Which are CI/CD pitfalls for AI apps? (Select all that apply)
Using unit tests as the only gate (no eval gate)Deploying the 'latest' tag instead of a pinned SHACommitting secrets into the repoTagging images by immutable git SHA
Select all that apply.
Question 12medium
What kind of gate (two words) — running the golden-set regression suite in CI — blocks a merge on model-quality regression?
Question 13hard
What deploy strategy (one word) routes a small % of traffic to the new version and watches metrics before ramping up?

0 / 13 answered

CI/CD for AI Apps — Final Exam