Topic 03 Practice Test 2 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 Bedrock knowledge base using semantic chunking is producing too many tiny chunks from technical manuals. The team wants fewer, larger semantic chunks without abandoning semantic boundary detection. Which design choice best satisfies this requirement? Choose ONE.
- Set semantic chunking bufferSize to 1
- Lower semantic chunking maxTokens
- Increase the semantic breakpoint percentile threshold
- Use fixed-size chunking with overlap
Correct Answer: C
Correct Answer
Answer C is correct because This option is correct because it raises the breakpoint threshold so fewer semantic breaks qualify. That directly addresses semantic chunking splits coherent sections too often. Because semantic chunking splits coherent sections too often, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option uses one neighboring sentence on each side for boundary context. It fits when short sentences need adjacent context before boundaries are chosen. Here the decisive condition is semantic chunking splits coherent sections too often. In contrast, Increase the semantic breakpoint percentile threshold addresses that condition directly.
Answer B is incorrect because This option caps semantic chunks at a smaller maximum token count. It fits when meaningful sections still exceed the downstream context budget. Here the decisive condition is semantic chunking splits coherent sections too often. In contrast, Increase the semantic breakpoint percentile threshold addresses that condition directly.
Answer D is incorrect because This option creates predictable token windows with overlap between adjacent chunks. It fits when regular documents favor deterministic segmentation. Here the decisive condition is semantic chunking splits coherent sections too often. In contrast, Increase the semantic breakpoint percentile threshold addresses that condition directly.
Question 2
Many sentences in policy documents are short and ambiguous unless neighboring sentences are considered when deciding semantic boundaries. The team wants boundary detection to use adjacent sentence context. Which design choice best satisfies this requirement? Choose ONE.
- Increase the semantic breakpoint percentile threshold
- Use the Bedrock default chunking strategy
- Use semantic chunking
- Set semantic chunking bufferSize to 1
Correct Answer: D
Correct Answer
Answer D is correct because This option is correct because it uses one neighboring sentence on each side for boundary context. That directly addresses short sentences need adjacent context before boundaries are chosen. Because short sentences need adjacent context before boundaries are chosen, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option raises the breakpoint threshold so fewer semantic breaks qualify. It fits when semantic chunking splits coherent sections too often. Here the decisive condition is short sentences need adjacent context before boundaries are chosen. In contrast, Set semantic chunking bufferSize to 1 addresses that condition directly.
Answer B is incorrect because This option uses Bedrock fixed chunks of about 300 tokens with overlap. It fits when the managed default segmentation is sufficient. Here the decisive condition is short sentences need adjacent context before boundaries are chosen. In contrast, Set semantic chunking bufferSize to 1 addresses that condition directly.
Answer C is incorrect because This option places boundaries at semantic changes rather than fixed token positions. It fits when uneven document sections should be split by meaning. Here the decisive condition is short sentences need adjacent context before boundaries are chosen. In contrast, Set semantic chunking bufferSize to 1 addresses that condition directly.
Question 3
Semantic chunking is preserving meaning correctly, but some chunks are still too large for a downstream context window. The team wants to keep semantic segmentation while enforcing a smaller maximum chunk size. Which design choice best satisfies this requirement? Choose ONE.
- Increase the semantic breakpoint percentile threshold
- Tune the retrieval result count
- Use fixed-size chunking with overlap
- Lower semantic chunking maxTokens
Correct Answer: D
Correct Answer
Answer D is correct because This option is correct because it caps semantic chunks at a smaller maximum token count. That directly addresses meaningful sections still exceed the downstream context budget. Because meaningful sections still exceed the downstream context budget, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option raises the breakpoint threshold so fewer semantic breaks qualify. It fits when semantic chunking splits coherent sections too often. Here the decisive condition is meaningful sections still exceed the downstream context budget. In contrast, Lower semantic chunking maxTokens addresses that condition directly.
Answer B is incorrect because This option changes how many source chunks retrieval returns. It fits when the context set is too small or too large. Here the decisive condition is meaningful sections still exceed the downstream context budget. In contrast, Lower semantic chunking maxTokens addresses that condition directly.
Answer C is incorrect because This option creates predictable token windows with overlap between adjacent chunks. It fits when regular documents favor deterministic segmentation. Here the decisive condition is meaningful sections still exceed the downstream context budget. In contrast, Lower semantic chunking maxTokens addresses that condition directly.
Question 4
A team does not need a custom segmentation policy and wants the standard Bedrock Knowledge Bases behavior when no chunking configuration is supplied. Which design choice best satisfies this requirement? Choose ONE.
- Use semantic chunking
- Use the Bedrock default chunking strategy
- Use NONE chunking with a custom transformation Lambda
- Use a chunking strategy instead of NONE for page-aware retrieval
Correct Answer: B
Correct Answer
Answer B is correct because This option is correct because it uses Bedrock fixed chunks of about 300 tokens with overlap. That directly addresses the managed default segmentation is sufficient. Because the managed default segmentation is sufficient, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option places boundaries at semantic changes rather than fixed token positions. It fits when uneven document sections should be split by meaning. Here the decisive condition is the managed default segmentation is sufficient. In contrast, Use the Bedrock default chunking strategy addresses that condition directly.
Answer C is incorrect because This option lets transformation code emit proprietary chunks instead of native ones. It fits when custom source markers must define exact retrieval units. Here the decisive condition is the managed default segmentation is sufficient. In contrast, Use the Bedrock default chunking strategy addresses that condition directly.
Answer D is incorrect because This option uses actual chunks so page-aware retrieval metadata remains available. It fits when the application requires page-level citations or filters. Here the decisive condition is the managed default segmentation is sufficient. In contrast, Use the Bedrock default chunking strategy addresses that condition directly.
Question 5
An existing connected S3 data source was created with fixed-size chunking. The team now needs semantic chunking for the same content and discovers the original chunking strategy cannot be changed. Which design choice best satisfies this requirement? Choose ONE.
- Increase the semantic breakpoint percentile threshold
- Create a new data source with the required chunking strategy
- Use NONE chunking with a custom transformation Lambda
- Add a new data source with the new parsing strategy type
Correct Answer: B
Correct Answer
Answer B is correct because This option is correct because it moves content to a newly connected source with the desired chunker. That directly addresses a connected source must change its chunking strategy. Because a connected source must change its chunking strategy, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option raises the breakpoint threshold so fewer semantic breaks qualify. It fits when semantic chunking splits coherent sections too often. Here the decisive condition is a connected source must change its chunking strategy. In contrast, Create a new data source with the required chunking strategy addresses that condition directly.
Answer C is incorrect because This option lets transformation code emit proprietary chunks instead of native ones. It fits when custom source markers must define exact retrieval units. Here the decisive condition is a connected source must change its chunking strategy. In contrast, Create a new data source with the required chunking strategy addresses that condition directly.
Answer D is incorrect because This option uses a new source because parser strategy type is immutable after connection. It fits when an existing source must switch to a different parser type. Here the decisive condition is a connected source must change its chunking strategy. In contrast, Create a new data source with the required chunking strategy addresses that condition directly.
Question 6
A publisher uses proprietary XML markers to define retrieval units. None of the managed chunking strategies reproduce those boundaries, so the team wants its own Lambda code to emit the final chunks. Which design choice best satisfies this requirement? Choose ONE.
- Keep native chunking and add chunk-level metadata in the transformation Lambda
- Use the Bedrock default chunking strategy
- Use NONE chunking with a custom transformation Lambda
- Use semantic chunking
Correct Answer: C
Correct Answer
Answer C is correct because This option is correct because it lets transformation code emit proprietary chunks instead of native ones. That directly addresses custom source markers must define exact retrieval units. Because custom source markers must define exact retrieval units, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option keeps native chunks while enriching each chunk before indexing. It fits when derived product or version labels belong on individual chunks. Here the decisive condition is custom source markers must define exact retrieval units. In contrast, Use NONE chunking with a custom transformation Lambda addresses that condition directly.
Answer B is incorrect because This option uses Bedrock fixed chunks of about 300 tokens with overlap. It fits when the managed default segmentation is sufficient. Here the decisive condition is custom source markers must define exact retrieval units. In contrast, Use NONE chunking with a custom transformation Lambda addresses that condition directly.
Answer D is incorrect because This option places boundaries at semantic changes rather than fixed token positions. It fits when uneven document sections should be split by meaning. Here the decisive condition is custom source markers must define exact retrieval units. In contrast, Use NONE chunking with a custom transformation Lambda addresses that condition directly.
Question 7
Built-in chunking already produces good retrieval units, but each chunk needs a product-version label derived from its text before indexing. The team wants to preserve native chunking. Which design choice best satisfies this requirement? Choose ONE.
- Use NONE chunking with a custom transformation Lambda
- Apply an explicit metadata filter
- Preserve source URI and citation metadata
- Keep native chunking and add chunk-level metadata in the transformation Lambda
Correct Answer: D
Correct Answer
Answer D is correct because This option is correct because it keeps native chunks while enriching each chunk before indexing. That directly addresses derived product or version labels belong on individual chunks. Because derived product or version labels belong on individual chunks, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option lets transformation code emit proprietary chunks instead of native ones. It fits when custom source markers must define exact retrieval units. Here the decisive condition is derived product or version labels belong on individual chunks. In contrast, Keep native chunking and add chunk-level metadata in the transformation Lambda addresses that condition directly.
Answer B is incorrect because This option applies caller-supplied structured conditions before ranking candidates. It fits when the client already knows the exact filter values. Here the decisive condition is derived product or version labels belong on individual chunks. In contrast, Keep native chunking and add chunk-level metadata in the transformation Lambda addresses that condition directly.
Answer C is incorrect because This option preserves provenance so retrieved chunks can point to their source. It fits when users need traceable evidence links. Here the decisive condition is derived product or version labels belong on individual chunks. In contrast, Keep native chunking and add chunk-level metadata in the transformation Lambda addresses that condition directly.
Question 8
A user asks one request that compares three products, their regional restrictions, and current support status. Single-query retrieval often misses one part, so the team wants the request split into simpler retrieval subqueries. Which design choice best satisfies this requirement? Choose ONE.
- Enable query decomposition
- Apply query expansion before retrieval
- Use agentic retrieval
- Use Step Functions for a deterministic query-transformation workflow
Correct Answer: A
Correct Answer
Answer A is correct because This option is correct because it splits one compound request into simpler retrieval subqueries. That directly addresses a single request contains several independent information needs. Because a single request contains several independent information needs, this is the most direct design response.
Incorrect Answers
Answer B is incorrect because This option adds related terms or context before retrieval. It fits when terse acronym-heavy questions lack enough matching signals. Here the decisive condition is a single request contains several independent information needs. In contrast, Enable query decomposition addresses that condition directly.
Answer C is incorrect because This option plans iterative searches and continues until evidence is sufficient. It fits when a multi-hop question needs more than one planned retrieval step. Here the decisive condition is a single request contains several independent information needs. In contrast, Enable query decomposition addresses that condition directly.
Answer D is incorrect because This option orchestrates deterministic query normalization and routing with explicit workflow state. It fits when query rewrites require governed ordering and auditability. Here the decisive condition is a single request contains several independent information needs. In contrast, Enable query decomposition addresses that condition directly.
Question 9
A troubleshooting assistant must investigate multi-hop questions, retrieve evidence, decide whether it is sufficient, and issue additional searches until the answer is supported. Which design choice best satisfies this requirement? Choose ONE.
- Use agentic retrieval
- Enable query decomposition
- Call the Retrieve API
- Expose vector retrieval through foundation-model function calling
Correct Answer: A
Correct Answer
Answer A is correct because This option is correct because it plans iterative searches and continues until evidence is sufficient. That directly addresses a multi-hop question needs more than one planned retrieval step. Because a multi-hop question needs more than one planned retrieval step, this is the most direct design response.
Incorrect Answers
Answer B is incorrect because This option splits one compound request into simpler retrieval subqueries. It fits when a single request contains several independent information needs. Here the decisive condition is a multi-hop question needs more than one planned retrieval step. In contrast, Use agentic retrieval addresses that condition directly.
Answer C is incorrect because This option returns retrieved passages and metadata without generating the final answer. It fits when the application owns downstream ranking or generation. Here the decisive condition is a multi-hop question needs more than one planned retrieval step. In contrast, Use agentic retrieval addresses that condition directly.
Answer D is incorrect because This option exposes retrieval as a callable tool with a defined schema. It fits when the model should decide when evidence retrieval is needed. Here the decisive condition is a multi-hop question needs more than one planned retrieval step. In contrast, Use agentic retrieval addresses that condition directly.
Question 10
Users write requests such as ‘show only EU-approved procedures for finance’ but the client does not send structured filter fields. The team wants the retrieval layer to derive those metadata constraints from the request. Which design choice best satisfies this requirement? Choose ONE.
- Apply an explicit metadata filter
- Configure implicit filtering from the user query
- Apply query expansion before retrieval
- Store normalized domain-classification tags in metadata
Correct Answer: B
Correct Answer
Answer B is correct because This option is correct because it derives metadata restrictions from natural-language query context. That directly addresses users express filters in prose rather than structured fields. Because users express filters in prose rather than structured fields, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option applies caller-supplied structured conditions before ranking candidates. It fits when the client already knows the exact filter values. Here the decisive condition is users express filters in prose rather than structured fields. In contrast, Configure implicit filtering from the user query addresses that condition directly.
Answer C is incorrect because This option adds related terms or context before retrieval. It fits when terse acronym-heavy questions lack enough matching signals. Here the decisive condition is users express filters in prose rather than structured fields. In contrast, Configure implicit filtering from the user query addresses that condition directly.
Answer D is incorrect because This option normalizes business-domain labels across all ingested sources. It fits when source systems use inconsistent category names. Here the decisive condition is users express filters in prose rather than structured fields. In contrast, Configure implicit filtering from the user query addresses that condition directly.
Question 11
A custom application already owns answer generation and post-retrieval ranking. It needs Bedrock Knowledge Bases to return source chunks and metadata only, with no managed answer-generation step. Which design choice best satisfies this requirement? Choose ONE.
- Call the Retrieve API
- Use Amazon Bedrock Knowledge Bases managed RAG
- Use a custom reranker for a knowledge base with custom embeddings
- Publish a stable retrieval API contract between the FM layer and vector backend
Correct Answer: A
Correct Answer
Answer A is correct because This option is correct because it returns retrieved passages and metadata without generating the final answer. That directly addresses the application owns downstream ranking or generation. Because the application owns downstream ranking or generation, this is the most direct design response.
Incorrect Answers
Answer B is incorrect because This option manages ingestion retrieval and grounding as an integrated Bedrock workflow. It fits when the team wants a managed end-to-end RAG path. Here the decisive condition is the application owns downstream ranking or generation. In contrast, Call the Retrieve API addresses that condition directly.
Answer C is incorrect because This option applies a separately configured relevance model after retrieval. It fits when custom embeddings still require second-stage relevance scoring. Here the decisive condition is the application owns downstream ranking or generation. In contrast, Call the Retrieve API addresses that condition directly.
Answer D is incorrect because This option hides vector-store details behind a stable retrieval contract. It fits when applications must survive a later backend replacement. Here the decisive condition is the application owns downstream ranking or generation. In contrast, Call the Retrieve API addresses that condition directly.
Question 12
A shared S3 bucket contains approved knowledge under /published/ and unrelated staging files elsewhere. Only /published/ should ever enter the knowledge base. Which design choice best satisfies this requirement? Choose ONE.
- Apply an explicit metadata filter
- Store normalized domain-classification tags in metadata
- Use separate vector indexes for hard tenant isolation
- Set an S3 inclusion prefix for the approved path
Correct Answer: D
Correct Answer
Answer D is correct because This option is correct because it limits the connector to objects under an approved S3 prefix. That directly addresses only one governed folder belongs in the corpus. Because only one governed folder belongs in the corpus, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option applies caller-supplied structured conditions before ranking candidates. It fits when the client already knows the exact filter values. Here the decisive condition is only one governed folder belongs in the corpus. In contrast, Set an S3 inclusion prefix for the approved path addresses that condition directly.
Answer B is incorrect because This option normalizes business-domain labels across all ingested sources. It fits when source systems use inconsistent category names. Here the decisive condition is only one governed folder belongs in the corpus. In contrast, Set an S3 inclusion prefix for the approved path addresses that condition directly.
Answer C is incorrect because This option separates tenants into different vector indexes for storage-level isolation. It fits when policy forbids cross-tenant candidate visibility. Here the decisive condition is only one governed folder belongs in the corpus. In contrast, Set an S3 inclusion prefix for the approved path addresses that condition directly.
Question 13
A content platform must push or delete individual documents immediately through an API and attach metadata inline. It cannot wait for a connector synchronization job. Which design choice best satisfies this requirement? Choose ONE.
- Use a CUSTOM data source with KnowledgeBaseDocuments APIs
- Set an S3 inclusion prefix for the approved path
- Trigger the vector update pipeline from source-change events
- Mirror direct knowledge-base changes back to the S3 source of truth
Correct Answer: A
Correct Answer
Answer A is correct because This option is correct because it accepts direct document ingest and delete calls with metadata. That directly addresses documents must be updated immediately without connector synchronization. Because documents must be updated immediately without connector synchronization, this is the most direct design response.
Incorrect Answers
Answer B is incorrect because This option limits the connector to objects under an approved S3 prefix. It fits when only one governed folder belongs in the corpus. Here the decisive condition is documents must be updated immediately without connector synchronization. In contrast, Use a CUSTOM data source with KnowledgeBaseDocuments APIs addresses that condition directly.
Answer C is incorrect because This option starts vector updates when source-change events occur. It fits when new procedures must become searchable within minutes. Here the decisive condition is documents must be updated immediately without connector synchronization. In contrast, Use a CUSTOM data source with KnowledgeBaseDocuments APIs addresses that condition directly.
Answer D is incorrect because This option keeps direct knowledge-base edits consistent with the authoritative S3 objects. It fits when a later S3 sync must not restore stale content. Here the decisive condition is documents must be updated immediately without connector synchronization. In contrast, Use a CUSTOM data source with KnowledgeBaseDocuments APIs addresses that condition directly.
Question 14
Operators use direct ingestion to hot-fix a document in a knowledge base that is also connected to S3. S3 remains authoritative, and the next sync must not restore the old content. Which design choice best satisfies this requirement? Choose ONE.
- Use a CUSTOM data source with KnowledgeBaseDocuments APIs
- Mirror direct knowledge-base changes back to the S3 source of truth
- Use the metadata-only sync optimization
- Set the data-source deletion policy to RETAIN
Correct Answer: B
Correct Answer
Answer B is correct because This option is correct because it keeps direct knowledge-base edits consistent with the authoritative S3 objects. That directly addresses a later S3 sync must not restore stale content. Because a later S3 sync must not restore stale content, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option accepts direct document ingest and delete calls with metadata. It fits when documents must be updated immediately without connector synchronization. Here the decisive condition is a later S3 sync must not restore stale content. In contrast, Mirror direct knowledge-base changes back to the S3 source of truth addresses that condition directly.
Answer C is incorrect because This option reuses stored embeddings while updating eligible metadata-only changes. It fits when document content is unchanged and embedding calls should be avoided. Here the decisive condition is a later S3 sync must not restore stale content. In contrast, Mirror direct knowledge-base changes back to the S3 source of truth addresses that condition directly.
Answer D is incorrect because This option keeps indexed vector data after its source connection is deleted. It fits when the connection can disappear while converted knowledge must remain. Here the decisive condition is a later S3 sync must not restore stale content. In contrast, Mirror direct knowledge-base changes back to the S3 source of truth addresses that condition directly.
Question 15
Only .metadata.json files changed for non-CSV documents, there is no custom transformation Lambda, and the team wants to avoid another embedding-model call for unchanged content. Which design choice best satisfies this requirement? Choose ONE.
- Batch embedding generation through a Lambda-based ingestion workflow
- Re-embed the corpus with the new embedding model
- Use the metadata-only sync optimization
- Trigger the vector update pipeline from source-change events
Correct Answer: C
Correct Answer
Answer C is correct because This option is correct because it reuses stored embeddings while updating eligible metadata-only changes. That directly addresses document content is unchanged and embedding calls should be avoided. Because document content is unchanged and embedding calls should be avoided, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option groups large embedding workloads into controlled ingestion batches. It fits when millions of documents need efficient bulk embedding. Here the decisive condition is document content is unchanged and embedding calls should be avoided. In contrast, Use the metadata-only sync optimization addresses that condition directly.
Answer B is incorrect because This option regenerates stored vectors in the new embedding model’s representation space. It fits when an embedding-model migration changes vector geometry. Here the decisive condition is document content is unchanged and embedding calls should be avoided. In contrast, Use the metadata-only sync optimization addresses that condition directly.
Answer D is incorrect because This option starts vector updates when source-change events occur. It fits when new procedures must become searchable within minutes. Here the decisive condition is document content is unchanged and embedding calls should be avoided. In contrast, Use the metadata-only sync optimization addresses that condition directly.
Question 16
A data source currently uses a foundation-model parser. The architecture team wants to switch the source to Bedrock Data Automation, which is a different parsing strategy type. Which design choice best satisfies this requirement? Choose ONE.
- Create a new data source with the required chunking strategy
- Configure SupplementalDataStorageLocation when creating the knowledge base
- Add a new data source with the new parsing strategy type
- Use NONE chunking with a custom transformation Lambda
Correct Answer: C
Correct Answer
Answer C is correct because This option is correct because it uses a new source because parser strategy type is immutable after connection. That directly addresses an existing source must switch to a different parser type. Because an existing source must switch to a different parser type, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option moves content to a newly connected source with the desired chunker. It fits when a connected source must change its chunking strategy. Here the decisive condition is an existing source must switch to a different parser type. In contrast, Add a new data source with the new parsing strategy type addresses that condition directly.
Answer B is incorrect because This option reserves S3 storage for multimodal artifacts extracted during parsing. It fits when images or tables extracted by the parser must be retained. Here the decisive condition is an existing source must switch to a different parser type. In contrast, Add a new data source with the new parsing strategy type addresses that condition directly.
Answer D is incorrect because This option lets transformation code emit proprietary chunks instead of native ones. It fits when custom source markers must define exact retrieval units. Here the decisive condition is an existing source must switch to a different parser type. In contrast, Add a new data source with the new parsing strategy type addresses that condition directly.
Question 17
A new knowledge base will use a parser that extracts images, figures, charts, and tables. The extracted multimodal artifacts must be stored in S3 and available to retrieval. Which design choice best satisfies this requirement? Choose ONE.
- Set an S3 inclusion prefix for the approved path
- Configure SupplementalDataStorageLocation when creating the knowledge base
- Preserve source URI and citation metadata
- Configure the multimodal model’s audio or video chunk duration
Correct Answer: B
Correct Answer
Answer B is correct because This option is correct because it reserves S3 storage for multimodal artifacts extracted during parsing. That directly addresses images or tables extracted by the parser must be retained. Because images or tables extracted by the parser must be retained, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option limits the connector to objects under an approved S3 prefix. It fits when only one governed folder belongs in the corpus. Here the decisive condition is images or tables extracted by the parser must be retained. In contrast, Configure SupplementalDataStorageLocation when creating the knowledge base addresses that condition directly.
Answer C is incorrect because This option preserves provenance so retrieved chunks can point to their source. It fits when users need traceable evidence links. Here the decisive condition is images or tables extracted by the parser must be retained. In contrast, Configure SupplementalDataStorageLocation when creating the knowledge base addresses that condition directly.
Answer D is incorrect because This option sets temporal segments for native audio or video embeddings. It fits when retrieval quality depends on audio or video segment duration. Here the decisive condition is images or tables extracted by the parser must be retained. In contrast, Configure SupplementalDataStorageLocation when creating the knowledge base addresses that condition directly.
Question 18
A knowledge base uses native multimodal embeddings for recorded training sessions. Retrieval quality depends on temporal segmentation of audio and video rather than text-token chunk size. Which design choice best satisfies this requirement? Choose ONE.
- Lower semantic chunking maxTokens
- Use the Bedrock default chunking strategy
- Configure SupplementalDataStorageLocation when creating the knowledge base
- Configure the multimodal model’s audio or video chunk duration
Correct Answer: D
Correct Answer
Answer D is correct because This option is correct because it sets temporal segments for native audio or video embeddings. That directly addresses retrieval quality depends on audio or video segment duration. Because retrieval quality depends on audio or video segment duration, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option caps semantic chunks at a smaller maximum token count. It fits when meaningful sections still exceed the downstream context budget. Here the decisive condition is retrieval quality depends on audio or video segment duration. In contrast, Configure the multimodal model’s audio or video chunk duration addresses that condition directly.
Answer B is incorrect because This option uses Bedrock fixed chunks of about 300 tokens with overlap. It fits when the managed default segmentation is sufficient. Here the decisive condition is retrieval quality depends on audio or video segment duration. In contrast, Configure the multimodal model’s audio or video chunk duration addresses that condition directly.
Answer C is incorrect because This option reserves S3 storage for multimodal artifacts extracted during parsing. It fits when images or tables extracted by the parser must be retained. Here the decisive condition is retrieval quality depends on audio or video segment duration. In contrast, Configure the multimodal model’s audio or video chunk duration addresses that condition directly.
Question 19
A legacy source connection is being deleted, but the organization must keep the already converted and indexed vector data available for an audit period. Which design choice best satisfies this requirement? Choose ONE.
- Set the data-source deletion policy to RETAIN
- Mirror direct knowledge-base changes back to the S3 source of truth
- Use a CUSTOM data source with KnowledgeBaseDocuments APIs
- Use Amazon S3 Vectors for cost-focused, infrequent vector queries
Correct Answer: A
Correct Answer
Answer A is correct because This option is correct because it keeps indexed vector data after its source connection is deleted. That directly addresses the connection can disappear while converted knowledge must remain. Because the connection can disappear while converted knowledge must remain, this is the most direct design response.
Incorrect Answers
Answer B is incorrect because This option keeps direct knowledge-base edits consistent with the authoritative S3 objects. It fits when a later S3 sync must not restore stale content. Here the decisive condition is the connection can disappear while converted knowledge must remain. In contrast, Set the data-source deletion policy to RETAIN addresses that condition directly.
Answer C is incorrect because This option accepts direct document ingest and delete calls with metadata. It fits when documents must be updated immediately without connector synchronization. Here the decisive condition is the connection can disappear while converted knowledge must remain. In contrast, Set the data-source deletion policy to RETAIN addresses that condition directly.
Answer D is incorrect because This option provides durable low-cost vector storage for sporadic query traffic. It fits when storage cost dominates and sustained throughput is modest. Here the decisive condition is the connection can disappear while converted knowledge must remain. In contrast, Set the data-source deletion policy to RETAIN addresses that condition directly.
Question 20
An application requires page-aware citations and page-level filtering from PDF material. The current data source treats each document as one unchunked source unit. Which design choice best satisfies this requirement? Choose ONE.
- Use the Bedrock default chunking strategy
- Preserve source URI and citation metadata
- Use a chunking strategy instead of NONE for page-aware retrieval
- Call the Retrieve API
Correct Answer: C
Correct Answer
Answer C is correct because This option is correct because it uses actual chunks so page-aware retrieval metadata remains available. That directly addresses the application requires page-level citations or filters. Because the application requires page-level citations or filters, this is the most direct design response.
Incorrect Answers
Answer A is incorrect because This option uses Bedrock fixed chunks of about 300 tokens with overlap. It fits when the managed default segmentation is sufficient. Here the decisive condition is the application requires page-level citations or filters. In contrast, Use a chunking strategy instead of NONE for page-aware retrieval addresses that condition directly.
Answer B is incorrect because This option preserves provenance so retrieved chunks can point to their source. It fits when users need traceable evidence links. Here the decisive condition is the application requires page-level citations or filters. In contrast, Use a chunking strategy instead of NONE for page-aware retrieval addresses that condition directly.
Answer D is incorrect because This option returns retrieved passages and metadata without generating the final answer. It fits when the application owns downstream ranking or generation. Here the decisive condition is the application requires page-level citations or filters. In contrast, Use a chunking strategy instead of NONE for page-aware retrieval addresses that condition directly.
