Microsoft AI-102 (Designing and Implementing a Microsoft Azure AI Solution) Practice Test 54

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.

Question 531

Documents are embedded from one text-preprocessing pipeline but queries strip all product codes. What retrieval quality issue follows?

  1. Use the service's fused ranking and document identity handling rather than concatenating raw lists
  2. Update the integrated vectorizer or embedding skill connection to the correct model deployment
  3. Provide a useful initial candidate set because semantic ranker reranks retrieved results
  4. Make query and document preprocessing consistent for the intended retrieval task

Explanation

Correct Answer

D:

Differences in normalization and omitted identifiers can shift semantic representations and reduce recall. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after making query and document preprocessing consistent for the intended retrieval task.

Incorrect Answers

A:

Hybrid search ranks and merges multiple candidate sources; naive concatenation can duplicate one document in its own workflow, but that is a different correction from the present requirement to make query and document preprocessing consistent for the intended retrieval task. The reported issue would remain.

B:

Although useful elsewhere in the multilingual research-document retriever, vectorization depends on an accessible compatible model resource; stale deployment references break generation; the present case is different and calls for the team to make query and document preprocessing consistent for the intended retrieval task.

C:

That change addresses another failure mode: semantic ranking operates after initial retrieval rather than acting as a complete unbounded index crawler; it does not address this case, for which the relevant intervention is to make query and document preprocessing consistent for the intended retrieval task.

Question 532

A multilingual research-document retriever must rerank an initial candidate set using language-model relevance and return supported captions or answers. Reviewers will accept the implementation when returned hits include semantic reranking scores or captions when requested. Which approach best fulfills that requirement?

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

Explanation

Correct Answer

A:

Semantic ranker processes already retrieved text candidates and may extract captions and answer-like passages from them. For this multilingual research-document retriever, the intended evidence is that returned hits include semantic reranking scores or captions when requested. It does not create index entries or hallucinate new source facts as a generation model might.

Incorrect Answers

B:

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 rerank an initial candidate set using language-model relevance and return supported captions or answers. That separate capability does not prove that returned hits include semantic reranking scores or captions when requested; A keyword tokenizer alone does not create the numerical vector values, so selecting that alternative would not meet the stated acceptance check.

C:

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 rerank an initial candidate set using language-model relevance and return supported captions or answers. That separate capability does not prove that returned hits include semantic reranking scores or captions when requested; 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.

D:

Hybrid search executes both text and vector retrieval and fuses result rankings into a shared result set; it may be used to combine lexical matching with semantic vector retrieval for better coverage in a multilingual research-document retriever, but the current requirement is instead to rerank an initial candidate set using language-model relevance and return supported captions or answers. That separate capability does not prove that returned hits include semantic reranking scores or captions when requested; A vector-only request does not directly include a BM25 lexical matching branch, so selecting that alternative would not meet the stated acceptance check.

Question 533

An Azure AI Search solution must automatically generate retrievable vector chunks whenever new supported source documents are indexed. Which THREE pieces of the pipeline are needed? (Choose THREE.)

  1. Call a compatible embedding model to generate numeric vectors for those units
  2. Query a configured vector field with matching dimensions and vector-search settings
  3. Split supported source content into chunk-sized text units during enrichment
  4. Configure semantic search with eligible prioritized content fields and request semantic query behavior
  5. Submit text search and vector query components together in one supported Search request
  6. Map generated vectors and chunk identity into matching vector-enabled index fields

Explanation

Correct Answers

A:

Text alone does not become searchable as vectors; the indexing pipeline must obtain embeddings from a supported integration.

C:

Passage splitting creates retrievable sections and keeps long documents within practical model and index processing limits.

F:

An index needs schema-compatible vector fields and chunk associations before later nearest-neighbor queries can retrieve the ingested passages.

Incorrect Answers

B:

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 split supported source content into chunk-sized text units during enrichment and call a compatible embedding model to generate numeric vectors for those units and map generated vectors and chunk identity into matching vector-enabled index fields. The reported issue would remain.

D:

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 split supported source content into chunk-sized text units during enrichment and call a compatible embedding model to generate numeric vectors for those units and map generated vectors and chunk identity into matching vector-enabled index fields.

E:

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 split supported source content into chunk-sized text units during enrichment and call a compatible embedding model to generate numeric vectors for those units and map generated vectors and chunk identity into matching vector-enabled index fields.

Question 534

The application requests semantic captions but receives only ordinary search scores. What index configuration might be missing?

  1. Submit text search and vector query components together in one supported Search request
  2. Connect supported chunking and embedding skills with valid vector field mappings
  3. Configure semantic search with eligible prioritized content fields and request semantic query behavior
  4. Query a configured vector field with matching dimensions and vector-search settings

Explanation

Correct Answer

C:

Semantic reranking and captions require the relevant semantic configuration and query options. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after configuring semantic search with eligible prioritized content fields and request semantic query behavior.

Incorrect Answers

A:

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 configure semantic search with eligible prioritized content fields and request semantic query behavior.

B:

Integrated vectorization requires an indexing pipeline that generates and writes vector output, not OCR alone in its own workflow, but that is a different correction from the present requirement to configure semantic search with eligible prioritized content fields and request semantic query behavior. The reported issue would remain.

D:

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 configure semantic search with eligible prioritized content fields and request semantic query behavior.

Question 535

Indexing suddenly fails after a team replaces the embedding endpoint's deployment name. What should it verify?

  1. Update the integrated vectorizer or embedding skill connection to the correct model deployment
  2. Evaluate the configured vector similarity metric and nearest-neighbor behavior
  3. Provide a useful initial candidate set because semantic ranker reranks retrieved results
  4. Use the service's fused ranking and document identity handling rather than concatenating raw lists

Explanation

Correct Answer

A:

Vectorization depends on an accessible compatible model resource; stale deployment references break generation. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after updating the integrated vectorizer or embedding skill connection to the correct model deployment.

Incorrect Answers

B:

That change addresses another failure mode: vector search ranks relative similarity; it does not implement exact word equality or a universal semantic truth threshold; it does not address this case, for which the relevant intervention is to update the integrated vectorizer or embedding skill connection to the correct model deployment.

C:

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 update the integrated vectorizer or embedding skill connection to the correct model deployment. The reported issue would remain.

D:

Although useful elsewhere in the multilingual research-document retriever, hybrid search ranks and merges multiple candidate sources; naive concatenation can duplicate one document; the present case is different and calls for the team to update the integrated vectorizer or embedding skill connection to the correct model deployment.

Question 536

A vector request retrieves too few candidates for reranking. Which parameter or workflow stage should be reviewed?

  1. Evaluate both approaches on the same representative queries and relevance judgments
  2. Adjust the requested nearest-neighbor candidate count within supported limits
  3. Create passage-level embeddings and keep links back to the parent document
  4. Add a supported text-splitting skill and project child chunks appropriately

Explanation

Correct Answer

B:

Candidate breadth affects downstream reranking recall; a very small initial `k` can omit good documents. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after adjusting the requested nearest-neighbor candidate count within supported limits.

Incorrect Answers

A:

Although useful elsewhere in the multilingual research-document retriever, controlled comparisons reveal whether lexical recall complements vector similarity for the actual workload; the present case is different and calls for the team to adjust the requested nearest-neighbor candidate count within supported limits.

C:

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 adjust the requested nearest-neighbor candidate count within supported limits. 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 adjust the requested nearest-neighbor candidate count within supported limits.

Question 537

A developer expects semantic ranker to scan every document in a million-item corpus independently of initial retrieval. What architectural fact matters?

  1. Use the service's fused ranking and document identity handling rather than concatenating raw lists
  2. Provide a useful initial candidate set because semantic ranker reranks retrieved results
  3. Update the integrated vectorizer or embedding skill connection to the correct model deployment
  4. Evaluate the configured vector similarity metric and nearest-neighbor behavior

Explanation

Correct Answer

B:

Semantic ranking operates after initial retrieval rather than acting as a complete unbounded index crawler. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after providing a useful initial candidate set because semantic ranker reranks retrieved results.

Incorrect Answers

A:

Hybrid search ranks and merges multiple candidate sources; naive concatenation can duplicate one document in its own workflow, but that is a different correction from the present requirement to provide a useful initial candidate set because semantic ranker reranks retrieved results. The reported issue would remain.

C:

Although useful elsewhere in the multilingual research-document retriever, vectorization depends on an accessible compatible model resource; stale deployment references break generation; the present case is different and calls for the team to provide a useful initial candidate set because semantic ranker reranks retrieved results.

D:

That change addresses another failure mode: vector search ranks relative similarity; it does not implement exact word equality or a universal semantic truth threshold; it does not address this case, for which the relevant intervention is to provide a useful initial candidate set because semantic ranker reranks retrieved results.

Question 538

A release for a multilingual research-document retriever is blocked until the application can return index entries whose stored vectors are closest to the query vector. The tester will verify that the result set contains the most similar indexed vectors. What should be implemented?

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

Explanation

Correct Answer

B:

Vector search compares numeric representations according to the index's configured vector algorithms and search fields. For this multilingual research-document retriever, the intended evidence is that the result set contains the most similar indexed vectors. It does not require a literal exact word overlap between query and result.

Incorrect Answers

A:

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 return index entries whose stored vectors are closest to the query vector. That separate capability does not prove that the result set contains the most similar indexed vectors; A keyword tokenizer alone does not create the numerical vector values, 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 return index entries whose stored vectors are closest to the query vector. That separate capability does not prove that the result set contains the most similar indexed vectors; 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:

Hybrid search executes both text and vector retrieval and fuses result rankings into a shared result set; it may be used to combine lexical matching with semantic vector retrieval for better coverage in a multilingual research-document retriever, but the current requirement is instead to return index entries whose stored vectors are closest to the query vector. That separate capability does not prove that the result set contains the most similar indexed vectors; A vector-only request does not directly include a BM25 lexical matching branch, so selecting that alternative would not meet the stated acceptance check.

Question 539

A semantic search app finds irrelevant neighbors because it interprets vector distance as text equality. What should be tuned?

  1. Provide a useful initial candidate set because semantic ranker reranks retrieved results
  2. Update the integrated vectorizer or embedding skill connection to the correct model deployment
  3. Evaluate the configured vector similarity metric and nearest-neighbor behavior
  4. Make query and document preprocessing consistent for the intended retrieval task

Explanation

Correct Answer

C:

Vector search ranks relative similarity; it does not implement exact word equality or a universal semantic truth threshold. In the multilingual research-document retriever, confirm the fix by retesting the affected behavior after evaluating the configured vector similarity metric and nearest-neighbor behavior.

Incorrect Answers

A:

That change addresses another failure mode: semantic ranking operates after initial retrieval rather than acting as a complete unbounded index crawler; it does not address this case, for which the relevant intervention is to evaluate the configured vector similarity metric and nearest-neighbor behavior.

B:

Vectorization depends on an accessible compatible model resource; stale deployment references break generation in its own workflow, but that is a different correction from the present requirement to evaluate the configured vector similarity metric and nearest-neighbor behavior. The reported issue would remain.

D:

Although useful elsewhere in the multilingual research-document retriever, differences in normalization and omitted identifiers can shift semantic representations and reduce recall; the present case is different and calls for the team to evaluate the configured vector similarity metric and nearest-neighbor behavior.

Question 540

A team responsible for a multilingual research-document retriever needs to represent passages as numeric vectors suitable for semantic-neighbor retrieval; the acceptance check is that each selected passage has an embedding of the expected dimensions. Which capability should be selected?

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

Explanation

Correct Answer

B:

An embedding model encodes text into vector coordinates that preserve useful semantic relationships. For this multilingual research-document retriever, the intended evidence is that each selected passage has an embedding of the expected dimensions. A keyword tokenizer alone does not create the numerical vector values.

Incorrect Answers

A:

Hybrid search executes both text and vector retrieval and fuses result rankings into a shared result set; it may be used to combine lexical matching with semantic vector retrieval for better coverage in a multilingual research-document retriever, but the current requirement is instead to represent passages as numeric vectors suitable for semantic-neighbor retrieval. That separate capability does not prove that each selected passage has an embedding of the expected dimensions; A vector-only request does not directly include a BM25 lexical matching branch, 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 represent passages as numeric vectors suitable for semantic-neighbor retrieval. That separate capability does not prove that each selected passage has an embedding of the expected dimensions; 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:

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 represent passages as numeric vectors suitable for semantic-neighbor retrieval. That separate capability does not prove that each selected passage has an embedding of the expected dimensions; It does not require a literal exact word overlap between query and result, so selecting that alternative would not meet the stated acceptance check.

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