Microsoft AI-300 Online Practice Questions and Exam Preparation

Microsoft AI-300 Online Questions & Answers

• Question 11:

  An operations team provisions Azure Machine Learning workspaces repeatedly for new projects. The deployments must be automated, reviewed in source control, and reproducible.

  Which approach best meets the requirement?

  A. Deploy the workspace and related resources by using Bicep and Azure CLI from a GitHub Actions workflow
  B. Create each workspace manually in the Azure portal and export screenshots
  C. Store the workspace name in a notebook and run it during model training
  D. Create a prompt variant for each workspace configuration
  A. Deploy the workspace and related resources by using Bicep and Azure CLI from a GitHub Actions workflow

  ---

  Explanation

  Bicep, Azure CLI, and GitHub Actions provide an infrastructure-as-code workflow that is repeatable, reviewable, and automated from source control. Manual portal work is not reproducible, notebooks are not the right control plane for provisioning infrastructure, and prompt variants are unrelated to Azure Machine Learning resource deployment.

• Question 12:

  A team plans to deploy a large foundation model in Microsoft Foundry as part of a new enterprise AI capability.

  Different business units across the team's organization will access the model from various internal applications.

  You need to deploy a foundation model by minimizing latency.

  Which deployment type should you use?

  A. Developer
  B. Data Zone Batch
  C. Data Zone Standard
  D. Global Batch
  C. Data Zone Standard

  ---

  Explanation

  Provisioned throughput is appropriate when multiple business units require predictable capacity for a large foundation model in production. Serverless or ad hoc deployment may be easier to start with, but it does not reserve throughput for high-volume enterprise demand. The requirement focuses on production capacity planning rather than prompt design or data storage.

• Question 13:

  DRAG DROP

  A team manages prompts that are used by a generative AI application built on Microsoft Foundry. Multiple developers contribute prompt updates, and changes must be reviewed and tracked over time.

  The team requires that:

  1. Prompt changes are reviewed before being applied to the version in production.

  2. Previous prompt versions can be restored if issues occur.

  3. Prompt updates follow the same governance practices as the application code.

  You need to implement a controlled process for managing and updating prompts in production.

  How should you manage prompt updates to meet the requirements? To answer, move the appropriate actions to the correct requirements. You may use each action once, more than once, or not at all. You may need to move the split bar between panes or scroll to view content.

  NOTE: Each correct selection is worth one point.

  Select and Place:

  

  

  ---

  The image uses a Git repository for tracking prompt history and gated pull requests for review before production use. This gives prompt changes the same controlled source-control workflow as other application assets. Direct production edits or storage-only approaches would not provide the review trail and approval gate required by the scenario.

• Question 14:

  A RAG application often misses relevant passages even when those passages exist in the indexed content. The team wants to improve retrieval quality before changing the generation prompt.

  Which tuning area should the team review first?

  A. Similarity thresholds, chunk sizes, and retrieval strategy
  B. The number of archived model versions
  C. The managed identity display name
  D. The batch endpoint schedule
  A. Similarity thresholds, chunk sizes, and retrieval strategy

  ---

  Explanation

  RAG retrieval quality depends on settings such as similarity thresholds, chunk sizes, and retrieval strategy.

  Those settings determine which passages are retrieved before generation. Archived model versions, identity names, and batch endpoint schedules do not control retrieval recall. The requirement is to improve retrieval before changing the prompt.

• Question 15:

  A team is working in Microsoft Foundry to test and compare large language model (LLM) prompt variants in a development environment.

  The team requires consistent inputs to evaluate prompt variants without relying on live user traffic.

  You need to create a controlled evaluation of input data.

  Which action should you perform first?

  A. Generate synthetic interaction data.
  B. Configure content filters.
  C. Apply a blocklist.
  D. Enable observability metrics.
  A. Generate synthetic interaction data.

  ---

  Explanation

  Microsoft Foundry supports comparing prompt variants in a development workflow so teams can evaluate behavior before release. This directly matches prompt experimentation and performance comparison.

  Other choices such as production-only edits, storage containers, or endpoint configuration do not provide the focused prompt comparison capability required by the scenario.

• Question 16:

  HOTSPOT

  You use Azure Machine Learning to train models across multiple experiments by using the same workspace.

  You must record training runs in a centralized location to compare results from different jobs.

  During training, performance values must be captured so they appear in the experiment run history.

  You need to configure experiment tracking.

  What should you configure for each requirement? To answer, select the appropriate options in the answer area.

  NOTE: Each correct selection is worth one point.

  

  

  ---

  MLflow tracks for centralized run history and metric logging for training performance values. MLflow is the normal experiment tracking mechanism in Azure Machine Learning for parameters, metrics, and artifacts.

  Other items such as data assets, registries, or Azure Monitor workspaces can support adjacent operations, but they do not replace experiment tracking for run comparison.

• Question 17:

  A Machine Learning workspace must be accessible only through private network paths. The security team wants to reduce public exposure of the workspace.

  What should you configure?

  A. Private networking and restricted public network access for the workspace
  B. A larger compute cluster for training workloads
  C. A new model version in the registry
  D. A higher similarity threshold for retrieval
  A. Private networking and restricted public network access for the workspace

  ---

  Explanation

  Private networking and public access restrictions are the workspace controls that reduce exposure to the public internet. Compute size, model versioning, and retrieval thresholds address different operational needs. The requirement is a network security requirement, so the control must be applied at the Machine Learning workspace connectivity layer.

• Question 18:

  You create a binary classification model. You use the Fairlearn package to assess model fairness.

  You must eliminate the need to retrain the model.

  You need to implement the Fairlearn package.

  Which algorithm should you use?

  A. fairiearn.reductions.ExponentiatedGradient
  B. fairlearn.preprocessing.CorrelationRemover
  C. fairlearn.reductions.GridSearch
  D. fairlearn.postprocessing.ThresholdOptimizer
  D. fairlearn.postprocessing.ThresholdOptimizer

  ---

  Explanation

  Fairlearn postprocessing can adjust predictions to improve fairness without retraining the original binary classification model. Preprocessing and in-processing approaches generally affect training data or training algorithms, which would require retraining. The requirement is specifically to eliminate retraining, so a postprocessing mitigation is the right operational fit.

• Question 19:

  DRAG DROP

  A company plans to deploy a foundation model in Microsoft Foundry.

  The mode must support the following workloads:

  1. A customer support workload used across multiple regions

  2. A marketing workload that must remain within a specific region due to data residency requirements.

  You need to select the deployment type.

  Which deployment type should you use for each workload? To answer, move the appropriate deployment types to the correct requirements. You may use each deployment type once, more than once, or not at all.

  You may need to move the split bar between panes or scroll to view content.

  NOTE: Each correct selection is worth one point.

  Select and Place:

  

  

  ---

  The deployment image maps the multi-region customer support workload to a global deployment type and the data-residency marketing workload to a regional deployment type. Global deployment supports broad availability, while a regional deployment keeps processing within the chosen region. Other deployment choices would either over-constrain the global workload or violate the residency requirement.

• Question 20:

  A team runs training jobs that vary widely in compute demand. The team must avoid paying for idle nodes while keeping training jobs repeatable in Azure Machine Learning.

  Which compute target should the team configure?

  A. A compute cluster with autoscaling
  B. A fixed always-on virtual machine used by all jobs
  C. A Microsoft Foundry project environment
  D. A model registry
  A. A compute cluster with autoscaling

  ---

  Explanation

  An autoscaling Azure Machine Learning compute cluster can grow for training demand and shrink when jobs finish, reducing idle cost. Fixed virtual machines or dedicated always-on clusters are less cost-aware, and registries or Foundry project settings do not execute training jobs. The requirement combines repeatable training with variable capacity.