Microsoft AI-103 (Developing AI Apps and Agents on Azure) Practice Test 47

Question 460

A Foundry-based application is being scoped for production. The non-negotiable task is to ensure the index's vector field accepts embeddings produced by the selected model. Which implementation fits?

  1. Define searchable and filterable fields appropriately
  2. Match index vector dimensions to embedding model outputs
  3. Add semantic reranking for answer relevance
  4. Generate embeddings during indexing with a skillset

Explanation

Correct Answer

B:

For the stated requirement to ensure the index's vector field accepts embeddings produced by the selected model, an Azure AI Search vector field must have dimensions compatible with the embedding output and a valid vector search configuration; mismatched vectors fail at indexing or query time; this is the mechanism that the system must implement.

Incorrect Answers

A:

For a system whose objective is to support both full-text answers and exact filtering by department and policy year, Azure AI Search index fields must be configured for the operations used; searchable text differs from structured filterable metadata; this does not directly meet the required outcome to ensure the index's vector field accepts embeddings produced by the selected model.

C:

For a system whose objective is to promote passages that directly answer questions over textually similar but less useful chunks, semantic ranking reranks a first-stage candidate list using language understanding; it cannot rescue documents never retrieved; this does not directly meet the required outcome to ensure the index's vector field accepts embeddings produced by the selected model.

D:

For a system whose objective is to avoid writing a custom service for every supported text-to-vector conversion, integrated vectorization uses an indexer and enrichment pipeline to chunk and embed content; query-side vectorization still needs compatible settings; this does not directly meet the required outcome to ensure the index's vector field accepts embeddings produced by the selected model.

Question 461

Before approving the proposed action to project chunks to a secondary retrieval index, which measurable acceptance check should be required?

  1. Verify every returned chunk maps to the correct parent document and revision
  2. Check indexed vector coverage and embedding failures by document batch
  3. Measure ingestion lag and stale-citation rate following a controlled source update
  4. Compare indexer status and content versions with the source repository after updates

Explanation

Correct Answer

A:

The accepted change is to project chunks to a secondary retrieval index; this evaluation checks whether it can search fine-grained passages while preserving links to complete source documents, since index projections can place chunk records in a retrieval index with parent keys so answers can link back to source files.

Incorrect Answers

B:

This test concerns the ability to avoid writing a custom service for every supported text-to-vector conversion, for which integrated vectorization uses an indexer and enrichment pipeline to chunk and embed content; query-side vectorization still needs compatible settings; it does not establish whether the change to project chunks to a secondary retrieval index works for the current release.

C:

This test concerns the ability to avoid answering from superseded policy versions after files change, for which scheduled or change-aware ingestion and indexed version checks keep the retrieval corpus aligned with its authority; it does not establish whether the change to project chunks to a secondary retrieval index works for the current release.

D:

This test concerns the ability to keep search content synchronized with periodically changing files in a connected repository, for which Azure AI Search indexers read supported sources and drive enrichment and indexing on a schedule; external custom sources may need a push path; it does not establish whether the change to project chunks to a secondary retrieval index works for the current release.

Question 462

The team plans to attach retrieved source identifiers to agent tools. Which test provides the clearest evidence that this change works?

  1. Red-team searches with restricted accounts and audit unauthorized passage exposure
  2. Check table-cell relationships and heading attribution on representative document queries
  3. Check each quoted claim against the cited source record and revision
  4. Track retrieval recall at k separately from answer faithfulness and fluency
  5. Index sample vectors, issue test queries, and verify zero dimension or vector-configuration errors

Explanation

Correct Answer

C:

The accepted change is to attach retrieved source identifiers to agent tools; this evaluation checks whether it can make an agent's grounded answer traceable back to the correct document chunk, since a retrieval tool should return source identity and relevant snippet metadata along with content, not text alone.

Incorrect Answers

A:

This test concerns the ability to stop an agent from using search hits that the current requester is not allowed to read, for which permission-aware retrieval or correctly enforced security filters keep restricted chunks out of prompt context; it does not establish whether the change to attach retrieved source identifiers to agent tools works for the current release.

B:

This test concerns the ability to preserve relationships between headings, cells and surrounding text during document ingestion, for which layout processing detects document structure and table boundaries beyond simple plain-text OCR; it does not establish whether the change to attach retrieved source identifiers to agent tools works for the current release.

D:

This test concerns the ability to diagnose whether poor RAG answers originate from missing passages rather than generation itself, for which search relevance and recall metrics isolate retrieval defects; changing prompts cannot create facts absent from the context; it does not establish whether the change to attach retrieved source identifiers to agent tools works for the current release.

E:

This test concerns the ability to ensure the index's vector field accepts embeddings produced by the selected model, for which an Azure AI Search vector field must have dimensions compatible with the embedding output and a valid vector search configuration; mismatched vectors fail at indexing or query time; it does not establish whether the change to attach retrieved source identifiers to agent tools works for the current release.

Question 463

A Foundry-based application is being scoped for production. The non-negotiable task is to preserve relationships between headings, cells and surrounding text during document ingestion. Which implementation fits?

  1. Use layout-aware extraction for tables and headings
  2. Apply OCR to scanned document images
  3. Project chunks to a secondary retrieval index
  4. Add an image-description enrichment stage

Explanation

Correct Answer

A:

For the stated requirement to preserve relationships between headings, cells and surrounding text during document ingestion, layout processing detects document structure and table boundaries beyond simple plain-text OCR; this is the mechanism that the system must implement.

Incorrect Answers

B:

For a system whose objective is to make text from photographed forms available for retrieval and citation, OCR converts visible characters into machine-readable text; text splitters cannot recover words from image-only pages; this does not directly meet the required outcome to preserve relationships between headings, cells and surrounding text during document ingestion.

C:

For a system whose objective is to search fine-grained passages while preserving links to complete source documents, index projections can place chunk records in a retrieval index with parent keys so answers can link back to source files; this does not directly meet the required outcome to preserve relationships between headings, cells and surrounding text during document ingestion.

D:

For a system whose objective is to index meaningful visual content of diagrams that contains no searchable text, image captioning or multimodal understanding produces descriptors for visual semantics; OCR alone reads only visible characters; this does not directly meet the required outcome to preserve relationships between headings, cells and surrounding text during document ingestion.

Question 464

An engineering team has a requirement to avoid answering from superseded policy versions after files change. Which capability should be in the initial design?

  1. Attach retrieved source identifiers to agent tools
  2. Use freshness monitoring and incremental refresh
  3. Define searchable and filterable fields appropriately
  4. Filter retrieved evidence by user entitlements

Explanation

Correct Answer

B:

For the stated requirement to avoid answering from superseded policy versions after files change, scheduled or change-aware ingestion and indexed version checks keep the retrieval corpus aligned with its authority; this is the mechanism that the system must implement.

Incorrect Answers

A:

For a system whose objective is to make an agent's grounded answer traceable back to the correct document chunk, a retrieval tool should return source identity and relevant snippet metadata along with content, not text alone; this does not directly meet the required outcome to avoid answering from superseded policy versions after files change.

C:

For a system whose objective is to support both full-text answers and exact filtering by department and policy year, Azure AI Search index fields must be configured for the operations used; searchable text differs from structured filterable metadata; this does not directly meet the required outcome to avoid answering from superseded policy versions after files change.

D:

For a system whose objective is to stop an agent from using search hits that the current requester is not allowed to read, permission-aware retrieval or correctly enforced security filters keep restricted chunks out of prompt context; this does not directly meet the required outcome to avoid answering from superseded policy versions after files change.

Question 465

Before approving the proposed action to apply retrieval quality evaluation before model tuning, which measurable acceptance check should be required?

  1. Measure ingestion lag and stale-citation rate following a controlled source update
  2. Track retrieval recall at k separately from answer faithfulness and fluency
  3. Compare top-k answer relevance and reranker scores on labeled questions
  4. Red-team searches with restricted accounts and audit unauthorized passage exposure

Explanation

Correct Answer

B:

The accepted change is to apply retrieval quality evaluation before model tuning; this evaluation checks whether it can diagnose whether poor RAG answers originate from missing passages rather than generation itself, since search relevance and recall metrics isolate retrieval defects; changing prompts cannot create facts absent from the context.

Incorrect Answers

A:

This test concerns the ability to avoid answering from superseded policy versions after files change, for which scheduled or change-aware ingestion and indexed version checks keep the retrieval corpus aligned with its authority; it does not establish whether the change to apply retrieval quality evaluation before model tuning works for the current release.

C:

This test concerns the ability to promote passages that directly answer questions over textually similar but less useful chunks, for which semantic ranking reranks a first-stage candidate list using language understanding; it cannot rescue documents never retrieved; it does not establish whether the change to apply retrieval quality evaluation before model tuning works for the current release.

D:

This test concerns the ability to stop an agent from using search hits that the current requester is not allowed to read, for which permission-aware retrieval or correctly enforced security filters keep restricted chunks out of prompt context; it does not establish whether the change to apply retrieval quality evaluation before model tuning works for the current release.

Question 466

An Azure AI workload passes basic health probes but has this issue: A niche chemical code needs a company-specific normalization rule not supplied by a built-in skill. Which fix targets it?

  1. Add a supported OCR or document-reading enrichment step before chunking
  2. Implement a validated custom enrichment skill with clear input and output mappings
  3. Run layout-aware parsing and retain structural context in indexed passages
  4. Send one query containing lexical terms and vector search requests to the index

Explanation

Correct Answer

B:

The fault signal is that a niche chemical code needs a company-specific normalization rule not supplied by a built-in skill; the targeted fix works because custom Web API skills can call external enrichment code and return structured outputs to an indexer pipeline, restoring the ability to apply a proprietary extraction algorithm that the built-in search skillset cannot express.

Incorrect Answers

A:

If the failing function were to make text from photographed forms available for retrieval and citation, this change could be relevant because OCR converts visible characters into machine-readable text; text splitters cannot recover words from image-only pages; the actual symptom is that a niche chemical code needs a company-specific normalization rule not supplied by a built-in skill, which calls for a different repair.

C:

If the failing function were to preserve relationships between headings, cells and surrounding text during document ingestion, this change could be relevant because layout processing detects document structure and table boundaries beyond simple plain-text OCR; the actual symptom is that a niche chemical code needs a company-specific normalization rule not supplied by a built-in skill, which calls for a different repair.

D:

If the failing function were to rank both exact internal policy codes and semantically related explanations, this change could be relevant because hybrid retrieval combines text search and vector retrieval, merging candidates using reciprocal-rank fusion; the actual symptom is that a niche chemical code needs a company-specific normalization rule not supplied by a built-in skill, which calls for a different repair.

Question 467

A pilot has a functional gap: A public HR assistant retrieves executive compensation passages because it uses an all-access search credential. How should the team address that gap?

  1. Use an appropriate text split strategy with contextual metadata and modest overlap
  2. Expose source IDs and versioned citations in the retrieval tool response contract
  3. Enforce effective user permissions in retrieval queries and document mappings
  4. Refresh changed documents and verify version stamps before serving critical policy answers

Explanation

Correct Answer

C:

The fault signal is that a public HR assistant retrieves executive compensation passages because it uses an all-access search credential; the targeted fix works because permission-aware retrieval or correctly enforced security filters keep restricted chunks out of prompt context, restoring the ability to stop an agent from using search hits that the current requester is not allowed to read.

Incorrect Answers

A:

If the failing function were to keep retrieval passages within embedding and prompt size limits without losing section context, this change could be relevant because chunking creates smaller units for indexing and grounding; titles and parent identifiers should accompany chunks; the actual symptom is that a public HR assistant retrieves executive compensation passages because it uses an all-access search credential, which calls for a different repair.

B:

If the failing function were to make an agent's grounded answer traceable back to the correct document chunk, this change could be relevant because a retrieval tool should return source identity and relevant snippet metadata along with content, not text alone; the actual symptom is that a public HR assistant retrieves executive compensation passages because it uses an all-access search credential, which calls for a different repair.

D:

If the failing function were to avoid answering from superseded policy versions after files change, this change could be relevant because scheduled or change-aware ingestion and indexed version checks keep the retrieval corpus aligned with its authority; the actual symptom is that a public HR assistant retrieves executive compensation passages because it uses an all-access search credential, which calls for a different repair.

Question 468

An Azure AI workload passes basic health probes but has this issue: New documents become searchable by keywords but their vector fields are empty. Which fix targets it?

  1. Configure chunk-level index projections with parent identifiers and source metadata
  2. Add the appropriate embedding skill and index projections to the ingestion pipeline
  3. Use an appropriate text split strategy with contextual metadata and modest overlap
  4. Align the embedding deployment and vector field schema, then regenerate incompatible vectors

Explanation

Correct Answer

B:

The fault signal is that new documents become searchable by keywords but their vector fields are empty; the targeted fix works because integrated vectorization uses an indexer and enrichment pipeline to chunk and embed content; query-side vectorization still needs compatible settings, restoring the ability to avoid writing a custom service for every supported text-to-vector conversion.

Incorrect Answers

A:

If the failing function were to search fine-grained passages while preserving links to complete source documents, this change could be relevant because index projections can place chunk records in a retrieval index with parent keys so answers can link back to source files; the actual symptom is that new documents become searchable by keywords but their vector fields are empty, which calls for a different repair.

C:

If the failing function were to keep retrieval passages within embedding and prompt size limits without losing section context, this change could be relevant because chunking creates smaller units for indexing and grounding; titles and parent identifiers should accompany chunks; the actual symptom is that new documents become searchable by keywords but their vector fields are empty, which calls for a different repair.

D:

If the failing function were to ensure the index's vector field accepts embeddings produced by the selected model, this change could be relevant because an Azure AI Search vector field must have dimensions compatible with the embedding output and a valid vector search configuration; mismatched vectors fail at indexing or query time; the actual symptom is that new documents become searchable by keywords but their vector fields are empty, which calls for a different repair.

Question 469

After deployment, the team observes: Exact-code queries work, but natural-language variations fail; switching entirely to vectors would harm identifier searches. Which remediation addresses the likely cause?

  1. Add appropriately typed filterable metadata fields and populate them during indexing
  2. Align the embedding deployment and vector field schema, then regenerate incompatible vectors
  3. Send one query containing lexical terms and vector search requests to the index
  4. Refresh changed documents and verify version stamps before serving critical policy answers

Explanation

Correct Answer

C:

The fault signal is that exact-code queries work, but natural-language variations fail; switching entirely to vectors would harm identifier searches; the targeted fix works because hybrid retrieval combines text search and vector retrieval, merging candidates using reciprocal-rank fusion, restoring the ability to rank both exact internal policy codes and semantically related explanations.

Incorrect Answers

A:

If the failing function were to support both full-text answers and exact filtering by department and policy year, this change could be relevant because Azure AI Search index fields must be configured for the operations used; searchable text differs from structured filterable metadata; the actual symptom is that exact-code queries work, but natural-language variations fail; switching entirely to vectors would harm identifier searches, which calls for a different repair.

B:

If the failing function were to ensure the index's vector field accepts embeddings produced by the selected model, this change could be relevant because an Azure AI Search vector field must have dimensions compatible with the embedding output and a valid vector search configuration; mismatched vectors fail at indexing or query time; the actual symptom is that exact-code queries work, but natural-language variations fail; switching entirely to vectors would harm identifier searches, which calls for a different repair.

D:

If the failing function were to avoid answering from superseded policy versions after files change, this change could be relevant because scheduled or change-aware ingestion and indexed version checks keep the retrieval corpus aligned with its authority; the actual symptom is that exact-code queries work, but natural-language variations fail; switching entirely to vectors would harm identifier searches, which calls for a different repair.

Question 470

The engineering change request is to generate embeddings during indexing with a skillset. What is the most direct acceptance criterion?

  1. Measure answer-bearing chunk recall and index-size inflation after splitting
  2. Check each quoted claim against the cited source record and revision
  3. Check indexed vector coverage and embedding failures by document batch
  4. Verify every returned chunk maps to the correct parent document and revision

Explanation

Correct Answer

C:

The accepted change is to generate embeddings during indexing with a skillset; this evaluation checks whether it can avoid writing a custom service for every supported text-to-vector conversion, since integrated vectorization uses an indexer and enrichment pipeline to chunk and embed content; query-side vectorization still needs compatible settings.

Incorrect Answers

A:

This test concerns the ability to keep retrieval passages within embedding and prompt size limits without losing section context, for which chunking creates smaller units for indexing and grounding; titles and parent identifiers should accompany chunks; it does not establish whether the change to generate embeddings during indexing with a skillset works for the current release.

B:

This test concerns the ability to make an agent's grounded answer traceable back to the correct document chunk, for which a retrieval tool should return source identity and relevant snippet metadata along with content, not text alone; it does not establish whether the change to generate embeddings during indexing with a skillset works for the current release.

D:

This test concerns the ability to search fine-grained passages while preserving links to complete source documents, for which index projections can place chunk records in a retrieval index with parent keys so answers can link back to source files; it does not establish whether the change to generate embeddings during indexing with a skillset works for the current release.

Developing AI Apps and Agents on Azure (AI-103) Practice Tests