Free Exam AI-300: Operationalizing Machine Learning and Generative AI Solutions Practice Questions

Sample Practice Questions

Try these 10 sample questions from the Exam AI-300: Operationalizing Machine Learning and Generative AI Solutions practice bank. Each includes the answer and explanation. Continue with the interactive practice above to review the full question set.

1. 1An organization needs project-level cost reporting and separate datastore configurations for two Azure Machine Learning projects. Which workspace arrangement aligns with Microsoft's guidance?

   1. A. Use one workspace for each project.
   2. B. Use one experiment for each project in one workspace.
   3. C. Use one compute target for each project in one workspace.
   4. D. Use one registered model for each project in one workspace.

   Answer: A

   Explanation: Microsoft states, "limiting it to one project per workspace allows for cost reporting accrued to a project level." Separate workspaces also scope datastore configuration to each project; experiments, compute targets, and registered models do not provide this workspace boundary.

   Reference: What is an Azure Machine Learning workspace? — Organizing workspaces — Create a workspace per project

2. 2Which MLflow call records numeric validation accuracy at a particular training step?

   1. A. mlflow.log_param('accuracy', value)
   2. B. mlflow.set_experiment('accuracy')
   3. C. mlflow.log_artifact('accuracy.csv')
   4. D. mlflow.log_metric('accuracy', value, step=epoch)

   Answer: D

   Explanation: Microsoft states, "Use parameter step to indicate the step at which you log the metric value." log_metric records numeric measurements over time; parameters, experiment selection, and file artifacts do not record that metric series.

   Reference: Log metrics, parameters, and files with MLflow — Log metrics — numeric value over time

3. 3Within one centrally governed Foundry resource, which organizational boundary separates teams' development files, agents and evaluations while letting them reuse shared model deployments?

   1. A. A separate model deployment name for each team
   2. B. A separate Foundry project for each team
   3. C. A separate shared-service connection for each team
   4. D. A separate evaluation run name for each team

   Answer: B

   Explanation: "Development boundary inside the Foundry resource where teams build and evaluate use cases." Projects are development and artifact-access boundaries under the resource. Deployment names, connections and individual evaluation runs do not provide the same container for a team's files, agents and evaluations.

   Reference: Microsoft Foundry architecture — Resource hierarchy — Project

4. 4Which Foundry evaluator measures whether an answer's ideas are logically consistent and flow together?

   1. A. Fluency
   2. B. Relevance
   3. C. Groundedness
   4. D. Coherence

   Answer: D

   Explanation: Microsoft describes Coherence as "Measures logical consistency and flow of responses." Fluency concerns readability, relevance concerns the query, and groundedness concerns support in retrieved context.

   Reference: Built-in evaluators reference — General purpose evaluators — Coherence

5. 5An embedding request fails because a source document exceeds the model's token input limit. Which indexing change directly addresses that problem?

   1. A. Split the document into chunks within the model's input limit.
   2. B. Increase only the search query's top result count.
   3. C. Raise only the vector similarity threshold.
   4. D. Change only the final answer's response length.

   Answer: A

   Explanation: Microsoft states, "Partitioning large documents into smaller chunks can help you stay under the maximum token input limits of chat completion and embedding models." Search-result counts, thresholds, and answer limits do not shorten the oversized embedding input.

   Reference: Chunk large documents for RAG and vector search in Azure AI Search — Introduction — input token limits

6. 6An administrator must grant a team access to several Azure Machine Learning workspaces, preserve individual user identities, and avoid assigning each member separately to every workspace. What should the administrator use?

   1. A. Assign each user an individual role in every workspace.
   2. B. Assign workspace roles to a Microsoft Entra user group.
   3. C. Share one administrator account among all team members.
   4. D. Give every member the same workspace storage account key.

   Answer: B

   Explanation: "By using Microsoft Entra user groups, you don't have to add individual users to each workspace." Group role assignments avoid per-user repetition. Individual assignments preserve the administrative burden; a shared administrator account loses individual identity, and storage keys do not grant workspace membership.

   Reference: What is an Azure Machine Learning workspace? — Organizing workspaces — Assign access to user groups

7. 7A training script must record its chosen number of epochs as an MLflow experiment parameter. Which call is appropriate?

   1. A. mlflow.log_param('num_epochs', 20)
   2. B. mlflow.log_metric('num_epochs', 20)
   3. C. mlflow.set_tag('num_epochs', 20)
   4. D. mlflow.log_dict({'num_epochs': 20}, 'settings.json')

   Answer: A

   Explanation: "MLflow supports the logging parameters used by your experiments." Use log_param for the epoch configuration. log_metric records a measurement, set_tag records metadata, and log_dict stores a file artifact instead of a parameter.

   Reference: Log metrics, parameters, and files with MLflow — Log parameters — introductory paragraph and log_param example

8. 8A Foundry project connects to Azure Storage and Azure AI Search. How should their networking and access policies be governed?

   1. A. They automatically become identical to project policies.
   2. B. They are replaced by the model deployment's settings.
   3. C. They no longer require Azure resource permissions.
   4. D. They must be managed independently on the connected resources.

   Answer: D

   Explanation: Microsoft states, "These resources have separate governance boundaries, so you manage their networking and access policies independently." A connection does not transfer or replace the connected service governance boundary.

   Reference: Microsoft Foundry architecture — Resource hierarchy — Connected resources

9. 9A generated answer is relevant and factually supported but awkwardly written. Which Foundry evaluator specifically measures its language quality and readability?

   1. A. Fluency
   2. B. Groundedness
   3. C. Document Retrieval
   4. D. Response Completeness

   Answer: A

   Explanation: Microsoft describes Fluency as "Measures natural language quality and readability." Grounding, retrieval accuracy, and completeness do not specifically measure how naturally the answer is written.

   Reference: Built-in evaluators reference — General purpose evaluators — Fluency

10. 10Fixed-size chunks break relevant context at boundaries. Which change can preserve continuity while retaining fixed-size chunking?

    1. A. Partition adjacent chunks so that they share no boundary text.
    2. B. Repeat only the document title at the start of each adjacent chunk.
    3. C. Repeat a modest boundary passage in both adjacent chunks.
    4. D. Insert gaps between chunks to reduce their total indexed text.

    Answer: C

    Explanation: "overlapping a small amount of text between chunks can help maintain continuity and context." Boundary overlap retains a small passage in both chunks and can preserve continuity. Disjoint partitions or gaps lose boundary context; repeating only a title provides document identity but does not retain the adjoining passage.

    Reference: Chunk large documents for RAG and vector search in Azure AI Search — Content overlap considerations