Microsoft AI-102 (Designing and Implementing a Microsoft Azure AI Solution) Practice Test 53

Question 520

In a staff-facing search results page, engineers must promote results according to business-relevant relevance signals in addition to ordinary text matching. The expected evidence is that ranking reflects the configured business relevance signals. Which implementation is the best match?

  1. Use a scoring profile or ranking configuration
  2. Use a full-text search query
  3. Specify result sorting
  4. Apply an OData filter

Explanation

Correct Answer

A:

Search scoring configuration adjusts ranking using supported field weights or functions for an index. For this staff-facing search results page, the intended evidence is that ranking reflects the configured business relevance signals. A scoring profile does not ingest new documents or create vector embeddings.

Incorrect Answers

B:

Full-text search analyzes text and ranks matches according to the service's query behavior and configuration; it may be used to find indexed documents matching requested words using searchable text fields in a staff-facing search results page, but the current requirement is instead to promote results according to business-relevant relevance signals in addition to ordinary text matching. That separate capability does not prove that ranking reflects the configured business relevance signals; A filter expression alone does not provide relevance-ranked keyword matching, so selecting that alternative would not meet the stated acceptance check.

C:

Order-by clauses sort the returned set on fields whose index definitions permit sorting; it may be used to order matching records by a sortable field such as revision date in a staff-facing search results page, but the current requirement is instead to promote results according to business-relevant relevance signals in addition to ordinary text matching. That separate capability does not prove that ranking reflects the configured business relevance signals; Sorting is distinct from enforcing permission checks on individual documents, so selecting that alternative would not meet the stated acceptance check.

D:

Filters evaluate supported expressions on fields marked filterable and constrain the result set; it may be used to restrict search results by exact metadata conditions such as region or publication state in a staff-facing search results page, but the current requirement is instead to promote results according to business-relevant relevance signals in addition to ordinary text matching. That separate capability does not prove that ranking reflects the configured business relevance signals; A filter is not a natural-language summarizer or a token similarity search, so selecting that alternative would not meet the stated acceptance check.

Question 521

An indexer creates embeddings of the whole document when the application needs paragraph retrieval. What pipeline change is useful?

  1. Add a supported text-splitting skill and project child chunks appropriately
  2. Adjust the requested nearest-neighbor candidate count within supported limits
  3. Treat captions as excerpts or summaries of retrieved document content, not independent legal conclusions
  4. Create passage-level embeddings and keep links back to the parent document

Explanation

Correct Answer

A:

Integrated chunking allows smaller searchable units with their own embedding vectors and parent relationships. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after adding a supported text-splitting skill and project child chunks appropriately.

Incorrect Answers

B:

That change addresses another failure mode: candidate breadth affects downstream reranking recall; a very small initial `k` can omit good documents; it does not address this case, for which the relevant intervention is to add a supported text-splitting skill and project child chunks appropriately.

C:

Although useful elsewhere in the multilingual research-document retriever, semantic captions help identify relevant source text; authoritative interpretation still depends on the underlying document and context; the present case is different and calls for the team to add a supported text-splitting skill and project child chunks appropriately.

D:

Local chunk vectors capture subtopics that a single document-average representation may obscure in its own workflow, but that is a different correction from the present requirement to add a supported text-splitting skill and project child chunks appropriately. The reported issue would remain.

Question 522

An index field accepts 1,536-length vectors, but a new embedding model returns a different dimension. What should the schema change consider?

  1. Query a configured vector field with matching dimensions and vector-search settings
  2. Configure semantic search with eligible prioritized content fields and request semantic query behavior
  3. Align vector field dimensions with the selected embedding model's output
  4. Submit text search and vector query components together in one supported Search request

Explanation

Correct Answer

C:

Vector field dimensionality is fixed by configuration; mismatched arrays cannot be indexed or compared correctly. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after aligning vector field dimensions with the selected embedding model's output.

Incorrect Answers

A:

Nearest-neighbor operations require vector-enabled fields, not arbitrary text columns in its own workflow, but that is a different correction from the present requirement to align vector field dimensions with the selected embedding model's output. The reported issue would remain.

B:

Although useful elsewhere in the multilingual research-document retriever, semantic reranking and captions require the relevant semantic configuration and query options; the present case is different and calls for the team to align vector field dimensions with the selected embedding model's output.

D:

That change addresses another failure mode: the service can fuse lexical and vector rankings in a hybrid query; independent concatenated lists do not use native fusion behavior; it does not address this case, for which the relevant intervention is to align vector field dimensions with the selected embedding model's output.

Question 523

In a multilingual research-document retriever, engineers must combine lexical matching with semantic vector retrieval for better coverage. The expected evidence is that results reflect both lexical and semantic retrieval paths. Which implementation is the best match?

  1. Configure integrated vectorization
  2. Run a vector nearest-neighbor query
  3. Enable semantic ranking
  4. Generate embeddings for indexed text
  5. Run a hybrid keyword and vector search

Explanation

Correct Answer

E:

Hybrid search executes both text and vector retrieval and fuses result rankings into a shared result set. For this multilingual research-document retriever, the intended evidence is that results reflect both lexical and semantic retrieval paths. A vector-only request does not directly include a BM25 lexical matching branch.

Incorrect Answers

A:

Integrated vectorization can use an indexer, skillset, and configured model resources to prepare vector fields during ingestion; it may be used to automate chunking and embedding of supported source content in the search indexing pipeline in a multilingual research-document retriever, but the current requirement is instead to combine lexical matching with semantic vector retrieval for better coverage. That separate capability does not prove that results reflect both lexical and semantic retrieval paths; It does not remove the need for an index schema, data source, or supported vector model integration, so selecting that alternative would not meet the stated acceptance check.

B:

Vector search compares numeric representations according to the index's configured vector algorithms and search fields; it may be used to return index entries whose stored vectors are closest to the query vector in a multilingual research-document retriever, but the current requirement is instead to combine lexical matching with semantic vector retrieval for better coverage. That separate capability does not prove that results reflect both lexical and semantic retrieval paths; It does not require a literal exact word overlap between query and result, so selecting that alternative would not meet the stated acceptance check.

C:

Semantic ranker processes already retrieved text candidates and may extract captions and answer-like passages from them; it may be used to rerank an initial candidate set using language-model relevance and return supported captions or answers in a multilingual research-document retriever, but the current requirement is instead to combine lexical matching with semantic vector retrieval for better coverage. That separate capability does not prove that results reflect both lexical and semantic retrieval paths; It does not create index entries or hallucinate new source facts as a generation model might, so selecting that alternative would not meet the stated acceptance check.

D:

An embedding model encodes text into vector coordinates that preserve useful semantic relationships; it may be used to represent passages as numeric vectors suitable for semantic-neighbor retrieval in a multilingual research-document retriever, but the current requirement is instead to combine lexical matching with semantic vector retrieval for better coverage. That separate capability does not prove that results reflect both lexical and semantic retrieval paths; A keyword tokenizer alone does not create the numerical vector values, so selecting that alternative would not meet the stated acceptance check.

Question 524

Semantic captions summarize passages but a user treats them as model-generated legal advice. What should the UI clarify?

  1. Evaluate both approaches on the same representative queries and relevance judgments
  2. Create passage-level embeddings and keep links back to the parent document
  3. Add a supported text-splitting skill and project child chunks appropriately
  4. Treat captions as excerpts or summaries of retrieved document content, not independent legal conclusions

Explanation

Correct Answer

D:

Semantic captions help identify relevant source text; authoritative interpretation still depends on the underlying document and context. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after treating captions as excerpts or summaries of retrieved document content, not independent legal conclusions.

Incorrect Answers

A:

Controlled comparisons reveal whether lexical recall complements vector similarity for the actual workload in its own workflow, but that is a different correction from the present requirement to treat captions as excerpts or summaries of retrieved document content, not independent legal conclusions. The reported issue would remain.

B:

That change addresses another failure mode: local chunk vectors capture subtopics that a single document-average representation may obscure; it does not address this case, for which the relevant intervention is to treat captions as excerpts or summaries of retrieved document content, not independent legal conclusions.

C:

Although useful elsewhere in the multilingual research-document retriever, integrated chunking allows smaller searchable units with their own embedding vectors and parent relationships; the present case is different and calls for the team to treat captions as excerpts or summaries of retrieved document content, not independent legal conclusions.

Question 525

A hybrid endpoint returns duplicates from the lexical and vector branches. What merge behavior should the client expect?

  1. Make query and document preprocessing consistent for the intended retrieval task
  2. Use the service's fused ranking and document identity handling rather than concatenating raw lists
  3. Evaluate the configured vector similarity metric and nearest-neighbor behavior
  4. Provide a useful initial candidate set because semantic ranker reranks retrieved results

Explanation

Correct Answer

B:

Hybrid search ranks and merges multiple candidate sources; naive concatenation can duplicate one document. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after using the service's fused ranking and document identity handling rather than concatenating raw lists.

Incorrect Answers

A:

That change addresses another failure mode: differences in normalization and omitted identifiers can shift semantic representations and reduce recall; it does not address this case, for which the relevant intervention is to use the service's fused ranking and document identity handling rather than concatenating raw lists.

C:

Although useful elsewhere in the multilingual research-document retriever, vector search ranks relative similarity; it does not implement exact word equality or a universal semantic truth threshold; the present case is different and calls for the team to use the service's fused ranking and document identity handling rather than concatenating raw lists.

D:

Semantic ranking operates after initial retrieval rather than acting as a complete unbounded index crawler in its own workflow, but that is a different correction from the present requirement to use the service's fused ranking and document identity handling rather than concatenating raw lists. The reported issue would remain.

Question 526

A long research report is represented by one embedding and narrow questions miss specific paragraphs. What ingestion change is useful?

  1. Evaluate both approaches on the same representative queries and relevance judgments
  2. Treat captions as excerpts or summaries of retrieved document content, not independent legal conclusions
  3. Adjust the requested nearest-neighbor candidate count within supported limits
  4. Create passage-level embeddings and keep links back to the parent document

Explanation

Correct Answer

D:

Local chunk vectors capture subtopics that a single document-average representation may obscure. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after creating passage-level embeddings and keep links back to the parent document.

Incorrect Answers

A:

That change addresses another failure mode: controlled comparisons reveal whether lexical recall complements vector similarity for the actual workload; it does not address this case, for which the relevant intervention is to create passage-level embeddings and keep links back to the parent document.

B:

Semantic captions help identify relevant source text; authoritative interpretation still depends on the underlying document and context in its own workflow, but that is a different correction from the present requirement to create passage-level embeddings and keep links back to the parent document. The reported issue would remain.

C:

Although useful elsewhere in the multilingual research-document retriever, candidate breadth affects downstream reranking recall; a very small initial `k` can omit good documents; the present case is different and calls for the team to create passage-level embeddings and keep links back to the parent document.

Question 527

A test team compares hybrid retrieval with vector-only retrieval on unrelated query sets. What should it change?

  1. Evaluate both approaches on the same representative queries and relevance judgments
  2. Adjust the requested nearest-neighbor candidate count within supported limits
  3. Treat captions as excerpts or summaries of retrieved document content, not independent legal conclusions
  4. Add a supported text-splitting skill and project child chunks appropriately

Explanation

Correct Answer

A:

Controlled comparisons reveal whether lexical recall complements vector similarity for the actual workload. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after evaluating both approaches on the same representative queries and relevance judgments.

Incorrect Answers

B:

Although useful elsewhere in the multilingual research-document retriever, candidate breadth affects downstream reranking recall; a very small initial `k` can omit good documents; the present case is different and calls for the team to evaluate both approaches on the same representative queries and relevance judgments.

C:

Semantic captions help identify relevant source text; authoritative interpretation still depends on the underlying document and context in its own workflow, but that is a different correction from the present requirement to evaluate both approaches on the same representative queries and relevance judgments. The reported issue would remain.

D:

That change addresses another failure mode: integrated chunking allows smaller searchable units with their own embedding vectors and parent relationships; it does not address this case, for which the relevant intervention is to evaluate both approaches on the same representative queries and relevance judgments.

Question 528

A vector query returns an error because the target field is a plain text string. What index schema is needed?

  1. Connect supported chunking and embedding skills with valid vector field mappings
  2. Align vector field dimensions with the selected embedding model's output
  3. Submit text search and vector query components together in one supported Search request
  4. Query a configured vector field with matching dimensions and vector-search settings

Explanation

Correct Answer

D:

Nearest-neighbor operations require vector-enabled fields, not arbitrary text columns. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after querying a configured vector field with matching dimensions and vector-search settings.

Incorrect Answers

A:

That change addresses another failure mode: integrated vectorization requires an indexing pipeline that generates and writes vector output, not OCR alone; it does not address this case, for which the relevant intervention is to query a configured vector field with matching dimensions and vector-search settings.

B:

Vector field dimensionality is fixed by configuration; mismatched arrays cannot be indexed or compared correctly in its own workflow, but that is a different correction from the present requirement to query a configured vector field with matching dimensions and vector-search settings. The reported issue would remain.

C:

Although useful elsewhere in the multilingual research-document retriever, the service can fuse lexical and vector rankings in a hybrid query; independent concatenated lists do not use native fusion behavior; the present case is different and calls for the team to query a configured vector field with matching dimensions and vector-search settings.

Question 529

The solution brief for a multilingual research-document retriever requires the system to automate chunking and embedding of supported source content in the search indexing pipeline. A successful test confirms that new source documents are vectorized as part of the indexing workflow. What should the team use?

  1. Run a vector nearest-neighbor query
  2. Configure integrated vectorization
  3. Generate embeddings for indexed text
  4. Enable semantic ranking

Explanation

Correct Answer

B:

Integrated vectorization can use an indexer, skillset, and configured model resources to prepare vector fields during ingestion. For this multilingual research-document retriever, the intended evidence is that new source documents are vectorized as part of the indexing workflow. It does not remove the need for an index schema, data source, or supported vector model integration.

Incorrect Answers

A:

Vector search compares numeric representations according to the index's configured vector algorithms and search fields; it may be used to return index entries whose stored vectors are closest to the query vector in a multilingual research-document retriever, but the current requirement is instead to automate chunking and embedding of supported source content in the search indexing pipeline. That separate capability does not prove that new source documents are vectorized as part of the indexing workflow; It does not require a literal exact word overlap between query and result, so selecting that alternative would not meet the stated acceptance check.

C:

An embedding model encodes text into vector coordinates that preserve useful semantic relationships; it may be used to represent passages as numeric vectors suitable for semantic-neighbor retrieval in a multilingual research-document retriever, but the current requirement is instead to automate chunking and embedding of supported source content in the search indexing pipeline. That separate capability does not prove that new source documents are vectorized as part of the indexing workflow; A keyword tokenizer alone does not create the numerical vector values, so selecting that alternative would not meet the stated acceptance check.

D:

Semantic ranker processes already retrieved text candidates and may extract captions and answer-like passages from them; it may be used to rerank an initial candidate set using language-model relevance and return supported captions or answers in a multilingual research-document retriever, but the current requirement is instead to automate chunking and embedding of supported source content in the search indexing pipeline. That separate capability does not prove that new source documents are vectorized as part of the indexing workflow; It does not create index entries or hallucinate new source facts as a generation model might, so selecting that alternative would not meet the stated acceptance check.

Question 530

A retrieval app sends separate lexical and vector requests and concatenates their first ten results. How should it construct a native hybrid query?

  1. Query a configured vector field with matching dimensions and vector-search settings
  2. Configure semantic search with eligible prioritized content fields and request semantic query behavior
  3. Submit text search and vector query components together in one supported Search request
  4. Align vector field dimensions with the selected embedding model's output

Explanation

Correct Answer

C:

The service can fuse lexical and vector rankings in a hybrid query; independent concatenated lists do not use native fusion behavior. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after submiting text search and vector query components together in one supported Search request.

Incorrect Answers

A:

That change addresses another failure mode: nearest-neighbor operations require vector-enabled fields, not arbitrary text columns; it does not address this case, for which the relevant intervention is to submit text search and vector query components together in one supported Search request.

B:

Semantic reranking and captions require the relevant semantic configuration and query options in its own workflow, but that is a different correction from the present requirement to submit text search and vector query components together in one supported Search request. The reported issue would remain.

D:

Although useful elsewhere in the multilingual research-document retriever, vector field dimensionality is fixed by configuration; mismatched arrays cannot be indexed or compared correctly; the present case is different and calls for the team to submit text search and vector query components together in one supported Search request.

Designing and Implementing a Microsoft Azure AI Solution (AI-102) Practice Tests