Topic 03 Practice Test 1 covers Vector Stores, Metadata, Chunking, Embeddings, Retrieval, RAG, and Data Maintenance in Domain 1 (Tasks 1.4-1.5). For broader exam preparation, review the AIP-C01 Exam Dumps. Every option includes focused technical reasoning tied to the tested decision.
Question 1
A infrastructure team must preserve meaning across uneven document sections. Which approach most directly satisfies the stated requirement? Choose ONE.
- tune top-k retrieval
- semantic chunking
- separate vector indexes for hard tenant isolation
- retrieval evaluation separate from generation evaluation
Correct Answer: B
Correct Answer
Answer B is correct because semantic chunking is correct because it splits by meaning to preserve coherent passages. The stem needs the team to preserve meaning across uneven document sections. That fit makes semantic chunking meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because tune top-k retrieval works by changes the number of nearest candidates returned so context is neither starved nor flooded. It fits where the model misses evidence with too few passages or receives distracting context with too many. The required function here is semantic chunking, so this option targets the wrong condition.
Answer C is incorrect because Use separate vector indexes for hard tenant isolation to create a storage boundary when policy forbids cross-tenant candidate visibility. Its mechanism places tenants in distinct vector collections or indexes rather than relying only on ranking behavior. The current scenario requires semantic chunking; separate vector indexes for hard tenant isolation solves a different design issue.
Answer D is incorrect because retrieval evaluation separate from generation evaluation works by measures retrieval recall/relevance independently before judging answer quality. It fits where the team must diagnose whether missing evidence or generation behavior causes poor RAG answers. The required function here is semantic chunking, so this option targets the wrong condition.
Question 2
A architecture review board must avoid losing facts that cross a mechanical chunk boundary. Which approach most directly satisfies the stated requirement? Choose ONE.
- minimum similarity threshold
- fixed-size chunking with overlap
- document normalization before embedding
- semantic chunking
Correct Answer: B
Correct Answer
Answer B is correct because fixed-size chunking with overlap is correct because it uses predictable token windows plus overlap so adjacent context survives segmentation. The stem needs the team to avoid losing facts that cross a mechanical chunk boundary. That fit makes fixed-size chunking with overlap meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because minimum similarity threshold works by requires candidate similarity to exceed a floor before a passage can enter retrieved context. It fits where low-confidence nearest neighbors are causing off-topic grounding. The required function here is fixed-size chunking with overlap, so this option targets the wrong condition.
Answer C is incorrect because document normalization before embedding works by cleans irrelevant headers, navigation text, or encoding noise before chunks are embedded. It fits where source documents contain repeated nonsemantic text that distorts vector similarity. The required function here is fixed-size chunking with overlap, so this option targets the wrong condition.
Answer D is incorrect because The role of semantic chunking is to preserve meaning across uneven document sections. It is useful when uneven sections require meaning-driven boundaries. Here the design needs fixed-size chunking with overlap; this choice instead solves an adjacent problem.
Question 3
A production operations group must exclude unauthorized departments before semantic similarity is evaluated. Which approach most directly satisfies the stated requirement? Choose ONE.
- metadata filtering before vector ranking
- delete stale vectors with deleted source content
- Amazon Bedrock Knowledge Bases
- fixed-size chunking with overlap
Correct Answer: A
Correct Answer
Answer A is correct because metadata filtering before vector ranking is correct because it applies structured metadata constraints to narrow the candidate set before vector ranking. The stem needs the team to exclude unauthorized departments before semantic similarity is evaluated. That fit makes metadata filtering before vector ranking meet the constraint; alternatives differ materially.
Incorrect Answers
Answer B is incorrect because delete stale vectors with deleted source content can be appropriate because it propagates source deletion to the vector store and associated metadata records. Apply it where content has been retired for legal, product, or policy reasons. This case depends on metadata filtering before vector ranking, leaving delete stale vectors with deleted source content mismatched to the requirement.
Answer C is incorrect because The role of Amazon Bedrock Knowledge Bases is to use a managed AWS RAG workflow for ingestion, retrieval, and FM grounding. It is useful when the team wants managed RAG integration instead of building every retrieval component itself. Here the design needs metadata filtering before vector ranking; this choice instead solves an adjacent problem.
Answer D is incorrect because The role of fixed-size chunking with overlap is to avoid losing facts that cross a mechanical chunk boundary. It is useful when documents are regular and deterministic chunk size is preferred. Here the design needs metadata filtering before vector ranking; this choice instead solves an adjacent problem.
Question 4
A platform team must keep query vectors compatible with vectors already stored in the index. Which approach most directly satisfies the stated requirement? Choose ONE.
- incremental ingestion for changed documents
- OpenSearch Serverless vector search
- consistent embedding dimensions
- metadata filtering before vector ranking
Correct Answer: C
Correct Answer
Answer C is correct because consistent embedding dimensions is correct because it uses the same embedding model and vector dimensionality on both query and corpus sides. The stem needs the team to keep query vectors compatible with vectors already stored in the index. That fit makes consistent embedding dimensions meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because The role of incremental ingestion for changed documents is to refresh only source material that changed rather than rebuilding everything. It is useful when a knowledge corpus changes continuously and full reindexing is unnecessarily expensive. Here the design needs consistent embedding dimensions; this choice instead solves an adjacent problem.
Answer B is incorrect because Use OpenSearch Serverless vector search to store and search embeddings with managed vector capabilities in OpenSearch Serverless. Its mechanism indexes vector fields and performs nearest-neighbor retrieval while retaining searchable metadata. The current scenario requires consistent embedding dimensions; OpenSearch Serverless vector search solves a different design issue.
Answer D is incorrect because The role of metadata filtering before vector ranking is to exclude unauthorized departments before semantic similarity is evaluated. It is useful when retrieval must respect tenant, region, product, or access labels. Here the design needs consistent embedding dimensions; this choice instead solves an adjacent problem.
Question 5
A network operations group must restore semantic compatibility after switching embedding models. Which approach most directly satisfies the stated requirement? Choose ONE.
- query-time metadata constraints
- task-appropriate embedding model selection
- re-embed corpus after model migration
- consistent embedding dimensions
Correct Answer: C
Correct Answer
Answer C is correct because re-embed corpus after model migration is correct because it regenerates stored vectors with the new embedding model rather than mixing old and new vector spaces. The stem needs the team to restore semantic compatibility after switching embedding models. That fit makes re-embed corpus after model migration meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because query-time metadata constraints can be appropriate because it passes request-specific metadata filters with the retrieval operation. Apply it where authorization or locale constraints differ for each user request. This case depends on re-embed corpus after model migration, leaving query-time metadata constraints mismatched to the requirement.
Answer B is incorrect because task-appropriate embedding model selection works by selects an embedding model based on supported languages, dimensions, quality, latency, and cost. It fits where retrieval quality is weak because the representation model does not fit the corpus or query characteristics. The required function here is re-embed corpus after model migration, so this option targets the wrong condition.
Answer D is incorrect because The role of consistent embedding dimensions is to keep query vectors compatible with vectors already stored in the index. It is useful when a vector index rejects or miscompares vectors from mismatched representations. Here the design needs re-embed corpus after model migration; this choice instead solves an adjacent problem.
Question 6
A application team must retain exact identifiers while also finding semantically related passages. Which approach most directly satisfies the stated requirement? Choose ONE.
- preserve source URI and citation metadata
- parent-child chunk mapping
- re-embed corpus after model migration
- hybrid keyword and vector retrieval
Correct Answer: D
Correct Answer
Answer D is correct because hybrid keyword and vector retrieval is correct because it combines lexical matching with semantic vector similarity and merges the candidate results. The stem needs the team to retain exact identifiers while also finding semantically related passages. That fit makes hybrid keyword and vector retrieval meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because preserve source URI and citation metadata can be appropriate because it stores source identifiers with chunks so retrieved passages can carry provenance into the response. Apply it where users need evidence links, auditability, or source attribution. This case depends on hybrid keyword and vector retrieval, leaving preserve source URI and citation metadata mismatched to the requirement.
Answer B is incorrect because parent-child chunk mapping belongs where small chunks retrieve accurately but lack explanatory context by themselves. It helps to retrieve precise child chunks while retaining enough parent context for generation. The stem instead requires hybrid keyword and vector retrieval, so parent-child chunk mapping does not meet the decisive condition.
Answer C is incorrect because re-embed corpus after model migration works by regenerates stored vectors with the new embedding model rather than mixing old and new vector spaces. It fits where an embedding migration changes representation geometry or dimensions. The required function here is hybrid keyword and vector retrieval, so this option targets the wrong condition.
Question 7
A enterprise architect must improve final relevance after a broad first-pass vector search. Which approach most directly satisfies the stated requirement? Choose ONE.
- separate vector indexes for hard tenant isolation
- retrieval evaluation separate from generation evaluation
- hybrid keyword and vector retrieval
- rerank retrieved candidates
Correct Answer: D
Correct Answer
Answer D is correct because rerank retrieved candidates is correct because it scores a smaller retrieved set with a stronger relevance model before context is sent to the FM. The stem needs the team to improve final relevance after a broad first-pass vector search. That fit makes rerank retrieved candidates meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because Use separate vector indexes for hard tenant isolation to create a storage boundary when policy forbids cross-tenant candidate visibility. Its mechanism places tenants in distinct vector collections or indexes rather than relying only on ranking behavior. The current scenario requires rerank retrieved candidates; separate vector indexes for hard tenant isolation solves a different design issue.
Answer B is incorrect because retrieval evaluation separate from generation evaluation works by measures retrieval recall/relevance independently before judging answer quality. It fits where the team must diagnose whether missing evidence or generation behavior causes poor RAG answers. The required function here is rerank retrieved candidates, so this option targets the wrong condition.
Answer C is incorrect because hybrid keyword and vector retrieval belongs where queries contain product codes, names, or exact terms alongside natural-language intent. It helps to retain exact identifiers while also finding semantically related passages. The stem instead requires rerank retrieved candidates, so hybrid keyword and vector retrieval does not meet the decisive condition.
Question 8
A site reliability team must control how many passages are supplied to the generation step. Which approach most directly satisfies the stated requirement? Choose ONE.
- document normalization before embedding
- tune top-k retrieval
- semantic chunking
- rerank retrieved candidates
Correct Answer: B
Correct Answer
Answer B is correct because tune top-k retrieval is correct because it changes the number of nearest candidates returned so context is neither starved nor flooded. The stem needs the team to control how many passages are supplied to the generation step. That fit makes tune top-k retrieval meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because document normalization before embedding works by cleans irrelevant headers, navigation text, or encoding noise before chunks are embedded. It fits where source documents contain repeated nonsemantic text that distorts vector similarity. The required function here is tune top-k retrieval, so this option targets the wrong condition.
Answer C is incorrect because The role of semantic chunking is to preserve meaning across uneven document sections. It is useful when uneven sections require meaning-driven boundaries. Here the design needs tune top-k retrieval; this choice instead solves an adjacent problem.
Answer D is incorrect because The role of rerank retrieved candidates is to improve final relevance after a broad first-pass vector search. It is useful when initial recall is good but top-ranked passages are not precise enough. Here the design needs tune top-k retrieval; this choice instead solves an adjacent problem.
Question 9
A security engineering group must drop weak semantic matches instead of forcing irrelevant context into the prompt. Which approach most directly satisfies the stated requirement? Choose ONE.
- Amazon Bedrock Knowledge Bases
- minimum similarity threshold
- fixed-size chunking with overlap
- tune top-k retrieval
Correct Answer: B
Correct Answer
Answer B is correct because minimum similarity threshold is correct because it requires candidate similarity to exceed a floor before a passage can enter retrieved context. The stem needs the team to drop weak semantic matches instead of forcing irrelevant context into the prompt. That fit makes minimum similarity threshold meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because The role of Amazon Bedrock Knowledge Bases is to use a managed AWS RAG workflow for ingestion, retrieval, and FM grounding. It is useful when the team wants managed RAG integration instead of building every retrieval component itself. Here the design needs minimum similarity threshold; this choice instead solves an adjacent problem.
Answer C is incorrect because The role of fixed-size chunking with overlap is to avoid losing facts that cross a mechanical chunk boundary. It is useful when documents are regular and deterministic chunk size is preferred. Here the design needs minimum similarity threshold; this choice instead solves an adjacent problem.
Answer D is incorrect because tune top-k retrieval can be appropriate because it changes the number of nearest candidates returned so context is neither starved nor flooded. Apply it where the model misses evidence with too few passages or receives distracting context with too many. This case depends on minimum similarity threshold, leaving tune top-k retrieval mismatched to the requirement.
Question 10
A branch deployment team must prevent removed documents from continuing to appear in RAG results. Which approach most directly satisfies the stated requirement? Choose ONE.
- OpenSearch Serverless vector search
- metadata filtering before vector ranking
- minimum similarity threshold
- delete stale vectors with deleted source content
Correct Answer: D
Correct Answer
Answer D is correct because delete stale vectors with deleted source content is correct because it propagates source deletion to the vector store and associated metadata records. The stem needs the team to prevent removed documents from continuing to appear in RAG results. That fit makes delete stale vectors with deleted source content meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because Use OpenSearch Serverless vector search to store and search embeddings with managed vector capabilities in OpenSearch Serverless. Its mechanism indexes vector fields and performs nearest-neighbor retrieval while retaining searchable metadata. The current scenario requires delete stale vectors with deleted source content; OpenSearch Serverless vector search solves a different design issue.
Answer B is incorrect because The role of metadata filtering before vector ranking is to exclude unauthorized departments before semantic similarity is evaluated. It is useful when retrieval must respect tenant, region, product, or access labels. Here the design needs delete stale vectors with deleted source content; this choice instead solves an adjacent problem.
Answer C is incorrect because minimum similarity threshold can be appropriate because it requires candidate similarity to exceed a floor before a passage can enter retrieved context. Apply it where low-confidence nearest neighbors are causing off-topic grounding. This case depends on delete stale vectors with deleted source content, leaving minimum similarity threshold mismatched to the requirement.
Question 11
A support organization must refresh only source material that changed rather than rebuilding everything. Which approach most directly satisfies the stated requirement? Choose ONE.
- task-appropriate embedding model selection
- consistent embedding dimensions
- incremental ingestion for changed documents
- delete stale vectors with deleted source content
Correct Answer: C
Correct Answer
Answer C is correct because incremental ingestion for changed documents is correct because it detects modified objects and updates their chunks and embeddings while preserving unaffected entries. The stem needs the team to refresh only source material that changed rather than rebuilding everything. That fit makes incremental ingestion for changed documents meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because task-appropriate embedding model selection works by selects an embedding model based on supported languages, dimensions, quality, latency, and cost. It fits where retrieval quality is weak because the representation model does not fit the corpus or query characteristics. The required function here is incremental ingestion for changed documents, so this option targets the wrong condition.
Answer B is incorrect because The role of consistent embedding dimensions is to keep query vectors compatible with vectors already stored in the index. It is useful when a vector index rejects or miscompares vectors from mismatched representations. Here the design needs incremental ingestion for changed documents; this choice instead solves an adjacent problem.
Answer D is incorrect because delete stale vectors with deleted source content belongs where content has been retired for legal, product, or policy reasons. It helps to prevent removed documents from continuing to appear in RAG results. The stem instead requires incremental ingestion for changed documents, so delete stale vectors with deleted source content does not meet the decisive condition.
Question 12
A cloud engineering team must apply dynamic user or session attributes to each retrieval request. Which approach most directly satisfies the stated requirement? Choose ONE.
- query-time metadata constraints
- parent-child chunk mapping
- re-embed corpus after model migration
- incremental ingestion for changed documents
Correct Answer: A
Correct Answer
Answer A is correct because query-time metadata constraints is correct because it passes request-specific metadata filters with the retrieval operation. The stem needs the team to apply dynamic user or session attributes to each retrieval request. That fit makes query-time metadata constraints meet the constraint; alternatives differ materially.
Incorrect Answers
Answer B is incorrect because parent-child chunk mapping belongs where small chunks retrieve accurately but lack explanatory context by themselves. It helps to retrieve precise child chunks while retaining enough parent context for generation. The stem instead requires query-time metadata constraints, so parent-child chunk mapping does not meet the decisive condition.
Answer C is incorrect because re-embed corpus after model migration works by regenerates stored vectors with the new embedding model rather than mixing old and new vector spaces. It fits where an embedding migration changes representation geometry or dimensions. The required function here is query-time metadata constraints, so this option targets the wrong condition.
Answer D is incorrect because Use incremental ingestion for changed documents to refresh only source material that changed rather than rebuilding everything. Its mechanism detects modified objects and updates their chunks and embeddings while preserving unaffected entries. The current scenario requires query-time metadata constraints; incremental ingestion for changed documents solves a different design issue.
Question 13
A infrastructure team must let generated answers trace evidence back to the originating document. Which approach most directly satisfies the stated requirement? Choose ONE.
- retrieval evaluation separate from generation evaluation
- hybrid keyword and vector retrieval
- preserve source URI and citation metadata
- query-time metadata constraints
Correct Answer: C
Correct Answer
Answer C is correct because preserve source URI and citation metadata is correct because it stores source identifiers with chunks so retrieved passages can carry provenance into the response. The stem needs the team to let generated answers trace evidence back to the originating document. That fit makes preserve source URI and citation metadata meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because retrieval evaluation separate from generation evaluation works by measures retrieval recall/relevance independently before judging answer quality. It fits where the team must diagnose whether missing evidence or generation behavior causes poor RAG answers. The required function here is preserve source URI and citation metadata, so this option targets the wrong condition.
Answer B is incorrect because hybrid keyword and vector retrieval belongs where queries contain product codes, names, or exact terms alongside natural-language intent. It helps to retain exact identifiers while also finding semantically related passages. The stem instead requires preserve source URI and citation metadata, so hybrid keyword and vector retrieval does not meet the decisive condition.
Answer D is incorrect because query-time metadata constraints belongs where authorization or locale constraints differ for each user request. It helps to apply dynamic user or session attributes to each retrieval request. The stem instead requires preserve source URI and citation metadata, so query-time metadata constraints does not meet the decisive condition.
Question 14
A architecture review board must create a storage boundary when policy forbids cross-tenant candidate visibility. Which approach most directly satisfies the stated requirement? Choose ONE.
- semantic chunking
- rerank retrieved candidates
- preserve source URI and citation metadata
- separate vector indexes for hard tenant isolation
Correct Answer: D
Correct Answer
Answer D is correct because separate vector indexes for hard tenant isolation is correct because it places tenants in distinct vector collections or indexes rather than relying only on ranking behavior. The stem needs the team to create a storage boundary when policy forbids cross-tenant candidate visibility. That fit makes separate vector indexes for hard tenant isolation meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because The role of semantic chunking is to preserve meaning across uneven document sections. It is useful when uneven sections require meaning-driven boundaries. Here the design needs separate vector indexes for hard tenant isolation; this choice instead solves an adjacent problem.
Answer B is incorrect because The role of rerank retrieved candidates is to improve final relevance after a broad first-pass vector search. It is useful when initial recall is good but top-ranked passages are not precise enough. Here the design needs separate vector indexes for hard tenant isolation; this choice instead solves an adjacent problem.
Answer C is incorrect because preserve source URI and citation metadata belongs where users need evidence links, auditability, or source attribution. It helps to let generated answers trace evidence back to the originating document. The stem instead requires separate vector indexes for hard tenant isolation, so preserve source URI and citation metadata does not meet the decisive condition.
Question 15
A production operations group must reduce retrieval noise caused by boilerplate markup and repeated formatting artifacts. Which approach most directly satisfies the stated requirement? Choose ONE.
- fixed-size chunking with overlap
- document normalization before embedding
- tune top-k retrieval
- separate vector indexes for hard tenant isolation
Correct Answer: B
Correct Answer
Answer B is correct because document normalization before embedding is correct because it cleans irrelevant headers, navigation text, or encoding noise before chunks are embedded. The stem needs the team to reduce retrieval noise caused by boilerplate markup and repeated formatting artifacts. That fit makes document normalization before embedding meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because The role of fixed-size chunking with overlap is to avoid losing facts that cross a mechanical chunk boundary. It is useful when documents are regular and deterministic chunk size is preferred. Here the design needs document normalization before embedding; this choice instead solves an adjacent problem.
Answer C is incorrect because tune top-k retrieval can be appropriate because it changes the number of nearest candidates returned so context is neither starved nor flooded. Apply it where the model misses evidence with too few passages or receives distracting context with too many. This case depends on document normalization before embedding, leaving tune top-k retrieval mismatched to the requirement.
Answer D is incorrect because separate vector indexes for hard tenant isolation is meant to create a storage boundary when policy forbids cross-tenant candidate visibility. Use it when regulatory or contractual isolation must be enforced below the application prompt layer. This stem calls for document normalization before embedding; separate vector indexes for hard tenant isolation addresses another operational need.
Question 16
A platform team must use a managed AWS RAG workflow for ingestion, retrieval, and FM grounding. Which approach most directly satisfies the stated requirement? Choose ONE.
- Amazon Bedrock Knowledge Bases
- metadata filtering before vector ranking
- minimum similarity threshold
- document normalization before embedding
Correct Answer: A
Correct Answer
Answer A is correct because Amazon Bedrock Knowledge Bases is correct because it coordinates supported data sources, chunking, embeddings, vector storage, and retrieval through a managed Bedrock feature. The stem needs the team to use a managed AWS RAG workflow for ingestion, retrieval, and FM grounding. That fit makes Amazon Bedrock Knowledge Bases meet the constraint; alternatives differ materially.
Incorrect Answers
Answer B is incorrect because The role of metadata filtering before vector ranking is to exclude unauthorized departments before semantic similarity is evaluated. It is useful when retrieval must respect tenant, region, product, or access labels. Here the design needs Amazon Bedrock Knowledge Bases; this choice instead solves an adjacent problem.
Answer C is incorrect because minimum similarity threshold can be appropriate because it requires candidate similarity to exceed a floor before a passage can enter retrieved context. Apply it where low-confidence nearest neighbors are causing off-topic grounding. This case depends on Amazon Bedrock Knowledge Bases, leaving minimum similarity threshold mismatched to the requirement.
Answer D is incorrect because document normalization before embedding can be appropriate because it cleans irrelevant headers, navigation text, or encoding noise before chunks are embedded. Apply it where source documents contain repeated nonsemantic text that distorts vector similarity. This case depends on Amazon Bedrock Knowledge Bases, leaving document normalization before embedding mismatched to the requirement.
Question 17
A network operations group must store and search embeddings with managed vector capabilities in OpenSearch Serverless. Which approach most directly satisfies the stated requirement? Choose ONE.
- OpenSearch Serverless vector search
- consistent embedding dimensions
- delete stale vectors with deleted source content
- Amazon Bedrock Knowledge Bases
Correct Answer: A
Correct Answer
Answer A is correct because OpenSearch Serverless vector search is correct because it indexes vector fields and performs nearest-neighbor retrieval while retaining searchable metadata. The stem needs the team to store and search embeddings with managed vector capabilities in OpenSearch Serverless. That fit makes OpenSearch Serverless vector search meet the constraint; alternatives differ materially.
Incorrect Answers
Answer B is incorrect because The role of consistent embedding dimensions is to keep query vectors compatible with vectors already stored in the index. It is useful when a vector index rejects or miscompares vectors from mismatched representations. Here the design needs OpenSearch Serverless vector search; this choice instead solves an adjacent problem.
Answer C is incorrect because delete stale vectors with deleted source content belongs where content has been retired for legal, product, or policy reasons. It helps to prevent removed documents from continuing to appear in RAG results. The stem instead requires OpenSearch Serverless vector search, so delete stale vectors with deleted source content does not meet the decisive condition.
Answer D is incorrect because Use Amazon Bedrock Knowledge Bases to use a managed AWS RAG workflow for ingestion, retrieval, and FM grounding. Its mechanism coordinates supported data sources, chunking, embeddings, vector storage, and retrieval through a managed Bedrock feature. The current scenario requires OpenSearch Serverless vector search; Amazon Bedrock Knowledge Bases solves a different design issue.
Question 18
A application team must choose embeddings that match the content language and retrieval behavior. Which approach most directly satisfies the stated requirement? Choose ONE.
- re-embed corpus after model migration
- incremental ingestion for changed documents
- task-appropriate embedding model selection
- OpenSearch Serverless vector search
Correct Answer: C
Correct Answer
Answer C is correct because task-appropriate embedding model selection is correct because it selects an embedding model based on supported languages, dimensions, quality, latency, and cost. The stem needs the team to choose embeddings that match the content language and retrieval behavior. That fit makes task-appropriate embedding model selection meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because re-embed corpus after model migration works by regenerates stored vectors with the new embedding model rather than mixing old and new vector spaces. It fits where an embedding migration changes representation geometry or dimensions. The required function here is task-appropriate embedding model selection, so this option targets the wrong condition.
Answer B is incorrect because Use incremental ingestion for changed documents to refresh only source material that changed rather than rebuilding everything. Its mechanism detects modified objects and updates their chunks and embeddings while preserving unaffected entries. The current scenario requires task-appropriate embedding model selection; incremental ingestion for changed documents solves a different design issue.
Answer D is incorrect because OpenSearch Serverless vector search is meant to store and search embeddings with managed vector capabilities in OpenSearch Serverless. Use it when the application needs managed vector search integrated with OpenSearch-style filtering. This stem calls for task-appropriate embedding model selection; OpenSearch Serverless vector search addresses another operational need.
Question 19
A enterprise architect must retrieve precise child chunks while retaining enough parent context for generation. Which approach most directly satisfies the stated requirement? Choose ONE.
- hybrid keyword and vector retrieval
- query-time metadata constraints
- task-appropriate embedding model selection
- parent-child chunk mapping
Correct Answer: D
Correct Answer
Answer D is correct because parent-child chunk mapping is correct because it links fine-grained retrieval units back to larger source sections so the FM receives coherent surrounding context. The stem needs the team to retrieve precise child chunks while retaining enough parent context for generation. That fit makes parent-child chunk mapping meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because hybrid keyword and vector retrieval belongs where queries contain product codes, names, or exact terms alongside natural-language intent. It helps to retain exact identifiers while also finding semantically related passages. The stem instead requires parent-child chunk mapping, so hybrid keyword and vector retrieval does not meet the decisive condition.
Answer B is incorrect because query-time metadata constraints belongs where authorization or locale constraints differ for each user request. It helps to apply dynamic user or session attributes to each retrieval request. The stem instead requires parent-child chunk mapping, so query-time metadata constraints does not meet the decisive condition.
Answer C is incorrect because task-appropriate embedding model selection can be appropriate because it selects an embedding model based on supported languages, dimensions, quality, latency, and cost. Apply it where retrieval quality is weak because the representation model does not fit the corpus or query characteristics. This case depends on parent-child chunk mapping, leaving task-appropriate embedding model selection mismatched to the requirement.
Question 20
A site reliability team must isolate whether failures come from search or from the FM response. Which approach most directly satisfies the stated requirement? Choose ONE.
- retrieval evaluation separate from generation evaluation
- rerank retrieved candidates
- preserve source URI and citation metadata
- parent-child chunk mapping
Correct Answer: A
Correct Answer
Answer A is correct because retrieval evaluation separate from generation evaluation is correct because it measures retrieval recall/relevance independently before judging answer quality. The stem needs the team to isolate whether failures come from search or from the FM response. That fit makes retrieval evaluation separate from generation evaluation meet the constraint; alternatives differ materially.
Incorrect Answers
Answer B is incorrect because The role of rerank retrieved candidates is to improve final relevance after a broad first-pass vector search. It is useful when initial recall is good but top-ranked passages are not precise enough. Here the design needs retrieval evaluation separate from generation evaluation; this choice instead solves an adjacent problem.
Answer C is incorrect because preserve source URI and citation metadata belongs where users need evidence links, auditability, or source attribution. It helps to let generated answers trace evidence back to the originating document. The stem instead requires retrieval evaluation separate from generation evaluation, so preserve source URI and citation metadata does not meet the decisive condition.
Answer D is incorrect because parent-child chunk mapping works by links fine-grained retrieval units back to larger source sections so the FM receives coherent surrounding context. It fits where small chunks retrieve accurately but lack explanatory context by themselves. The required function here is retrieval evaluation separate from generation evaluation, so this option targets the wrong condition.
