AI-300 cheat sheet

Updated September 28, 2026

Last-minute reference for AI-300. The model lifecycle domain is the largest; the three generative AI domains together are worth 40–55%.

Exam facts

Duration120 minutes
Passing score700 / 1000
PriceSet per country or region; shown at booking
LanguageEnglish only
ReplacesDP-100

Skill areas

AreaWeight
Implement machine learning model lifecycle and operations25–30%
Design and implement a GenAIOps infrastructure20–25%
Design and implement an MLOps infrastructure15–20%
Implement generative AI quality assurance and observability10–15%
Optimize generative AI systems and model performance10–15%

Requirement to answer

RequirementAnswer
Recreate a workspace identically per environmentBicep + Azure CLI
No stored Azure secret in GitHubOIDC federated credential
Approval before prod deploymentGitHub environment with required reviewers
Same model or environment in dev, test and prodRegistry
Read storage without keysIdentity-based datastore + managed identity
Training compute that costs nothing idleCompute cluster, min nodes 0
Track parameters and metrics with one linemlflow.autolog()
Search a continuous range and stop bad runsRandom sampling + bandit policy
Explain predictions, find failing cohortsResponsible AI dashboard
Real-time scoringManaged online endpoint
Score millions of files overnightBatch endpoint
Test on real traffic, no user impactMirrored traffic
Canary release and instant rollbackTraffic split between deployments
Alert when inputs changeModel monitor with data drift signal
App calls a model without a keyManaged identity + RBAC on Foundry
Steady high volume, predictable latency and costProvisioned throughput (PTUs)
Data must stay in a geographyRegional or data zone deployment, not global
Keep the tested model versionPinned version, no auto-upgrade
Answer contains facts not in the contextGroundedness
Answer doesn’t address the questionRelevance
Harmful content before launchRisk and safety evaluators, adversarial data
Why did this one request fail?Tracing in Application Insights
Exact codes missed by vector searchHybrid search
Consistent style or formatSupervised fine-tuning
Preferred vs rejected answer pairsDPO

Sweep sampling

SamplingContinuous valuesEarly termination
GridNoYes
RandomYesYes
BayesianYesNo

RAG levers

Chunk size and overlap, top-k, similarity threshold, hybrid search, semantic ranking, embedding model. Changing the embedding model means re-embedding everything.

Traps

  • A traffic split returns the new model’s answers to users. Mirroring does not.
  • Archiving a model version does not delete it.
  • Disabling public access breaks things unless storage, registry and DNS are private too.
  • Fluent and relevant is not grounded.
  • Evaluations that score everything zero usually mean a wrong column mapping.
  • Fine-tuning is not for facts that change. Use RAG.
  • Testing on synthetic data from the same run overstates performance.
  • DP-100 material covers the ML half and misses GenAIOps and IaC.

Night-before checklist

  • Name the five areas and which is largest
  • Online versus batch endpoint, traffic split versus mirroring, cold
  • The four quality metrics and what each measures
  • Check ID and proctoring rules; see exam day

Take the 20-question practice test and check your weakest area.