Topic 02 Practice Test 2 covers Multimodal Input Validation, Processing, Formatting, and Data Quality in Domain 1 (Task 1.3). For broader exam preparation, review the AIP-C01 Exam Dumps. Every option includes focused technical reasoning tied to the tested decision.
Question 1
A cloud engineering team must correct a data-processing weakness. An S3 table should contain one row per case_id, but upstream retries occasionally duplicate entire cases before summarization. Which solution most directly resolves this issue? Choose ONE.
- AWS Glue Data Quality completeness rule
- AWS Glue Data Quality uniqueness rule
- AWS Glue Data Quality numeric-range rule
- S3 quarantine prefix for failed validation
Correct Answer: B
Correct Answer
Answer B is correct because AWS Glue Data Quality uniqueness rule directly fits. It can measure duplicate keys in a managed ruleset and fail the quality check when uniqueness drops. The scenario is specifically about the case where duplicate business keys in a tabular batch would cause repeated evidence.
Incorrect Answers
Answer A is incorrect because AWS Glue Data Quality completeness rule helps when a batch table has mandatory fields and incomplete rows must be rejected by allowing the pipeline to evaluate required columns for nulls before records enter the FM pipeline. The described issue is duplicate business keys in a tabular batch would cause repeated evidence.
Answer C is incorrect because AWS Glue Data Quality numeric-range rule solves a different input problem. It helps to check numeric values against expected minimum and maximum boundaries as part of batch quality validation, normally when out-of-range measurements would mislead the FM. The current workflow needs duplicate business keys in a tabular batch would cause repeated evidence.
Answer D is incorrect because S3 quarantine prefix for failed validation is appropriate where invalid objects must be retained but prevented from reaching FM consumption. It achieves that by helping to isolate records that fail validation from the clean processing path while retaining them for investigation. This scenario instead requires duplicate business keys in a tabular batch would cause repeated evidence.
Question 2
A regulated workflow requires deterministic preprocessing before inference. Engineers received a new tabular corpus and do not yet know which columns have skew, missing values, or unexpected distributions. Which solution most directly resolves this issue? Choose ONE.
- deduplication by stable record identifier
- Lambda JSON Schema validation
- SageMaker Data Wrangler profiling
- Lambda file-signature validation
Correct Answer: C
Correct Answer
Answer C is correct because SageMaker Data Wrangler profiling should be used. It can profile distributions, missingness, and anomalies interactively before defining production transformations. That capability is needed because a team must understand data quality problems before finalizing preprocessing logic.
Incorrect Answers
Answer A is incorrect because deduplication by stable record identifier is designed to drop repeated records using a durable business key before constructing FM context. That makes sense when retries create duplicate records that would otherwise overweight the same evidence. The question is centered on a team must understand data quality problems before finalizing preprocessing logic.
Answer B is incorrect because Lambda JSON Schema validation can enforce the expected structure and data types for each event before invoking the model. Its best fit is where small event payloads need custom application-level validation at ingestion time. The current case instead needs a team must understand data quality problems before finalizing preprocessing logic.
Answer D is incorrect because Lambda file-signature validation is relevant, but it is intended to inspect magic bytes and declared content type so mislabeled or unsupported files are rejected early when uploaded files can have misleading extensions and must be validated before multimodal processing. The actual requirement is a team must understand data quality problems before finalizing preprocessing logic.
Question 3
A developer is hardening an FM-consumption workflow. Validation already rejects malformed records, but the operations team needs an alert if more than 3% of inputs fail during any 15-minute period. Which solution most directly resolves this issue? Choose ONE.
- SageMaker Data Wrangler profiling
- Lambda JSON Schema validation
- CloudWatch invalid-record-rate metric
- S3 quarantine prefix for failed validation
Correct Answer: C
Correct Answer
Answer C is correct because CloudWatch invalid-record-rate metric matches the constraint. It can publish a quality metric and alarm when the proportion of rejected inputs exceeds an operational threshold. This is the intended approach when operators need trend visibility into data-quality failures rather than only per-record rejection.
Incorrect Answers
Answer A is incorrect because SageMaker Data Wrangler profiling works for cases where a team must understand data quality problems before finalizing preprocessing logic by helping to profile distributions, missingness, and anomalies interactively before defining production transformations. The present scenario needs operators need trend visibility into data-quality failures rather than only per-record rejection.
Answer B is incorrect because Lambda JSON Schema validation targets a neighboring concern. It can enforce the expected structure and data types for each event before invoking the model when small event payloads need custom application-level validation at ingestion time. The key requirement here is operators need trend visibility into data-quality failures rather than only per-record rejection.
Answer D is incorrect because S3 quarantine prefix for failed validation can isolate records that fail validation from the clean processing path while retaining them for investigation. Its best fit is where invalid objects must be retained but prevented from reaching FM consumption. The current case instead needs operators need trend visibility into data-quality failures rather than only per-record rejection.
Question 4
An application team is standardizing data before foundation-model inference. A CSV feed contains invoice_total as a numeric value in most rows but sometimes includes nonnumeric text such as ‘unknown’. Which solution most directly resolves this issue? Choose ONE.
- tabular type coercion with explicit failure handling
- SageMaker Data Wrangler profiling
- Lambda file-signature validation
- checksum integrity verification
Correct Answer: A
Correct Answer
Answer A is correct because tabular type coercion with explicit failure handling is correct. It convert fields to expected types and route conversion failures to an error path instead of silently guessing. This matters because structured inputs contain type drift that must be made deterministic before FM formatting. That directly addresses the scenario.
Incorrect Answers
Answer B is incorrect because SageMaker Data Wrangler profiling would improve another stage by enabling it to profile distributions, missingness, and anomalies interactively before defining production transformations. It fits when a team must understand data quality problems before finalizing preprocessing logic. The stated problem is structured inputs contain type drift that must be made deterministic before FM formatting.
Answer C is incorrect because Lambda file-signature validation works for cases where uploaded files can have misleading extensions and must be validated before multimodal processing by helping to inspect magic bytes and declared content type so mislabeled or unsupported files are rejected early. The present scenario needs structured inputs contain type drift that must be made deterministic before FM formatting.
Answer D is incorrect because checksum integrity verification targets a neighboring concern. It can compare a supplied cryptographic checksum with the downloaded object before preprocessing when large artifacts can be corrupted in transit or during transfer and must be verified before use. The key requirement here is structured inputs contain type drift that must be made deterministic before FM formatting.
Question 5
A platform team is reviewing a production GenAI data pipeline. A document-processing API occasionally sends extremely long text that exceeds the selected model’s supported context window. Which solution most directly resolves this issue? Choose ONE.
- model input-size precheck
- Lambda JSON Schema validation
- required-field freshness validation
- cross-field consistency validation
Correct Answer: A
Correct Answer
Answer A is correct because model input-size precheck directly fits. It can estimate payload size or token count before invocation and reject or route oversized input. The scenario is specifically about the case where requests exceeding model limits should be handled deterministically before calling the FM.
Incorrect Answers
Answer B is incorrect because Lambda JSON Schema validation has value because it can enforce the expected structure and data types for each event before invoking the model. Its proper fit is when small event payloads need custom application-level validation at ingestion time. This question requires requests exceeding model limits should be handled deterministically before calling the FM.
Answer C is incorrect because required-field freshness validation can help a pipeline compare a source timestamp with an allowed staleness window before accepting the record, especially when the FM must operate only on sufficiently recent operational data. The present requirement is requests exceeding model limits should be handled deterministically before calling the FM, so it is not the best answer.
Answer D is incorrect because cross-field consistency validation would improve another stage by enabling it to evaluate related fields together so logically impossible combinations are rejected. It fits when data can be individually well-typed yet internally inconsistent. The stated problem is requests exceeding model limits should be handled deterministically before calling the FM.
Question 6
A cloud engineering team must correct a data-processing weakness. A nightly job must resize and normalize 500,000 images before they are sent to a multimodal evaluation workflow. Which solution most directly resolves this issue? Choose ONE.
- video key-frame extraction plus transcript pipeline
- SageMaker Processing batch media preprocessing
- large-file chunked preprocessing
- image dimension and format normalization
Correct Answer: B
Correct Answer
Answer B is correct because SageMaker Processing batch media preprocessing should be used. It can run scalable, repeatable preprocessing jobs over large media batches without tying work to request-time compute. That capability is needed because thousands of files need scheduled transformation before FM consumption.
Incorrect Answers
Answer A is incorrect because video key-frame extraction plus transcript pipeline is useful when full video is too large for direct use but both visual and spoken evidence matter because it can reduce a video to representative visual frames and speech text before multimodal reasoning. The current requirement is thousands of files need scheduled transformation before FM consumption, which makes this a mismatch.
Answer C is incorrect because Using large-file chunked preprocessing would let the team split a very large source artifact into bounded processing units before normalization and later reassembly. That is appropriate when a single preprocessing job would exceed memory or request limits. The scenario instead calls for thousands of files need scheduled transformation before FM consumption.
Answer D is incorrect because image dimension and format normalization has value because it can resize oversized images and convert them to a supported format before model invocation. Its proper fit is when input images exceed model/service constraints even though their visual content is valid. This question requires thousands of files need scheduled transformation before FM consumption.
Question 7
A regulated workflow requires deterministic preprocessing before inference. Recordings arrive as mono and stereo files with several sample rates. The speech-to-text stage expects a consistent supported audio format. Which solution most directly resolves this issue? Choose ONE.
- SageMaker Processing batch media preprocessing
- Step Functions multimodal orchestration
- audio normalization before transcription
- tabular rows serialized as structured JSON
Correct Answer: C
Correct Answer
Answer C is correct because audio normalization before transcription matches the constraint. It can standardize sample rate, channel layout, and encoding before speech recognition. This is the intended approach when heterogeneous audio encodings cause inconsistent transcription behavior.
Incorrect Answers
Answer A is incorrect because SageMaker Processing batch media preprocessing targets a different issue by helping to run scalable, repeatable preprocessing jobs over large media batches without tying work to request-time compute. It fits where thousands of files need scheduled transformation before FM consumption. The deciding condition is heterogeneous audio encodings cause inconsistent transcription behavior.
Answer B is incorrect because Step Functions multimodal orchestration can coordinate separate extraction, normalization, validation, and FM steps with explicit retries and failure branches. It fits when a multi-stage preprocessing workflow needs durable state and error handling. This scenario instead requires heterogeneous audio encodings cause inconsistent transcription behavior, so it targets a different problem.
Answer D is incorrect because tabular rows serialized as structured JSON addresses another concern by helping to convert selected columns into explicit key-value objects so the FM receives unambiguous field names and values. It fits where a tabular record must be represented clearly for inference rather than as an unlabeled value sequence. Here the deciding need is heterogeneous audio encodings cause inconsistent transcription behavior.
Question 8
A developer is hardening an FM-consumption workflow. Each uploaded text file should be normalized within seconds of arrival, and processing is short enough for a Lambda invocation. Which solution most directly resolves this issue? Choose ONE.
- SageMaker Processing batch media preprocessing
- language-aware transcription routing
- event-driven S3 Lambda preprocessing
- parallel modality preprocessing
Correct Answer: C
Correct Answer
Answer C is correct because event-driven S3 Lambda preprocessing is correct. It trigger lightweight preprocessing automatically when a new object lands in S3. This matters because small independent objects should begin processing immediately without a polling batch job. That directly addresses the scenario.
Incorrect Answers
Answer A is incorrect because SageMaker Processing batch media preprocessing belongs in workflows where thousands of files need scheduled transformation before FM consumption because it can run scalable, repeatable preprocessing jobs over large media batches without tying work to request-time compute. This question instead asks for small independent objects should begin processing immediately without a polling batch job.
Answer B is incorrect because language-aware transcription routing is useful when audio arrives in multiple languages and a single hard-coded language causes poor transcripts and can detect or configure the spoken language so the speech service uses the appropriate recognition path. The described workload requires small independent objects should begin processing immediately without a polling batch job.
Answer D is incorrect because parallel modality preprocessing can process independent audio, image, and text components concurrently before joining them for inference. It fits when multiple independent modalities create unnecessary latency when handled serially. This scenario instead requires small independent objects should begin processing immediately without a polling batch job, so it targets a different problem.
Question 9
An application team is standardizing data before foundation-model inference. A maintenance video must be summarized using both visible equipment states and spoken technician comments while minimizing irrelevant frames. Which solution most directly resolves this issue? Choose ONE.
- language-aware transcription routing
- tabular rows serialized as structured JSON
- table-header preservation during extraction
- video key-frame extraction plus transcript pipeline
Correct Answer: D
Correct Answer
Answer D is correct because video key-frame extraction plus transcript pipeline directly fits. It can reduce a video to representative visual frames and speech text before multimodal reasoning. The scenario is specifically about the case where full video is too large for direct use but both visual and spoken evidence matter.
Incorrect Answers
Answer A is incorrect because language-aware transcription routing would be sound if the goal were audio arrives in multiple languages and a single hard-coded language causes poor transcripts; it can detect or configure the spoken language so the speech service uses the appropriate recognition path. The requirement here is full video is too large for direct use but both visual and spoken evidence matter.
Answer B is incorrect because tabular rows serialized as structured JSON can convert selected columns into explicit key-value objects so the FM receives unambiguous field names and values, which matters when a tabular record must be represented clearly for inference rather than as an unlabeled value sequence. The present case is driven by full video is too large for direct use but both visual and spoken evidence matter.
Answer C is incorrect because table-header preservation during extraction belongs in workflows where separating values from their headers would destroy meaning needed by the FM because it can retain column headers with row values when converting tables to text or structured records. This question instead asks for full video is too large for direct use but both visual and spoken evidence matter.
Question 10
A platform team is reviewing a production GenAI data pipeline. Customer calls arrive in English, Spanish, and French. The pipeline must produce accurate text before the FM summarizes them. Which solution most directly resolves this issue? Choose ONE.
- Amazon Textract document extraction
- language-aware transcription routing
- Amazon Transcribe speech-to-text preprocessing
- audio normalization before transcription
Correct Answer: B
Correct Answer
Answer B is correct because language-aware transcription routing should be used. It can detect or configure the spoken language so the speech service uses the appropriate recognition path. That capability is needed because audio arrives in multiple languages and a single hard-coded language causes poor transcripts.
Incorrect Answers
Answer A is incorrect because Amazon Textract document extraction solves a different input problem. It helps to extract printed text, forms, and table structure from scanned documents before downstream language processing, normally when scanned forms need structured text and key-value information before an FM can reason over them. The current workflow needs audio arrives in multiple languages and a single hard-coded language causes poor transcripts.
Answer C is incorrect because Amazon Transcribe speech-to-text preprocessing would be sound if the goal were audio must become searchable and summarizable text before a text-only downstream step; it can convert spoken audio into text with timestamps before text-oriented FM processing. The requirement here is audio arrives in multiple languages and a single hard-coded language causes poor transcripts.
Answer D is incorrect because audio normalization before transcription can standardize sample rate, channel layout, and encoding before speech recognition, which matters when heterogeneous audio encodings cause inconsistent transcription behavior. The present case is driven by audio arrives in multiple languages and a single hard-coded language causes poor transcripts.
Question 11
A cloud engineering team must correct a data-processing weakness. A chat service wants to preserve user and assistant turns while reducing provider-specific message formatting code. Which solution most directly resolves this issue? Choose ONE.
- JSON escaping of untrusted text fields
- normalized UTF-8 text encoding
- Bedrock Converse message format
- batch-inference record envelope
Correct Answer: C
Correct Answer
Answer C is correct because Bedrock Converse message format matches the constraint. It can represent dialog turns with the standardized role/content message structure used by the Converse API. This is the intended approach when a multi-turn application needs a consistent conversational request format across supported models.
Incorrect Answers
Answer A is incorrect because JSON escaping of untrusted text fields helps when raw user text can otherwise produce malformed JSON request bodies by allowing the pipeline to use a serializer that escapes quotes, backslashes, and control characters instead of string concatenation. The described issue is a multi-turn application needs a consistent conversational request format across supported models.
Answer B is incorrect because normalized UTF-8 text encoding is relevant, but it is intended to convert incoming text to a consistent Unicode encoding before request serialization when mixed encodings create invalid characters and inconsistent tokenization. The actual requirement is a multi-turn application needs a consistent conversational request format across supported models.
Answer D is incorrect because batch-inference record envelope is appropriate where offline inference requires per-record association so outputs can be mapped back to source items. It achieves that by helping to emit each batch item using the service’s required record identifier and model-input envelope. This scenario instead requires a multi-turn application needs a consistent conversational request format across supported models.
Question 12
A regulated workflow requires deterministic preprocessing before inference. A generated prompt combines conversation history and retrieved passages and sometimes exceeds the model’s maximum input size. Which solution most directly resolves this issue? Choose ONE.
- Bedrock Converse message format
- speaker-label mapping to conversation roles
- system instruction separated from user content
- context-window-aware truncation before serialization
Correct Answer: D
Correct Answer
Answer D is correct because context-window-aware truncation before serialization is correct. It reduce or select input content before building the request so the final prompt fits the model context limit. This matters because request formatting must account for a bounded context window rather than failing at invocation. That directly addresses the scenario.
Incorrect Answers
Answer A is incorrect because Bedrock Converse message format is designed to represent dialog turns with the standardized role/content message structure used by the Converse API. That makes sense when a multi-turn application needs a consistent conversational request format across supported models. The question is centered on request formatting must account for a bounded context window rather than failing at invocation.
Answer B is incorrect because speaker-label mapping to conversation roles can map source labels such as AGENT and CUSTOMER into the model API’s supported conversational roles. Its best fit is where source-system speaker names are not directly valid message roles for the inference API. The current case instead needs request formatting must account for a bounded context window rather than failing at invocation.
Answer C is incorrect because system instruction separated from user content helps when the application must clearly distinguish global behavior from per-turn input by allowing the pipeline to place persistent behavioral instructions in the system field while keeping the user’s actual request in the user message. The described issue is request formatting must account for a bounded context window rather than failing at invocation.
Question 13
A developer is hardening an FM-consumption workflow. The input line ‘A17,EMEA,42.75’ must become an unambiguous structured record before the FM is asked to explain the transaction. Which solution most directly resolves this issue? Choose ONE.
- CSV-to-labeled-object transformation
- ordered conversation-turn serialization
- speaker-label mapping to conversation roles
- JSON escaping of untrusted text fields
Correct Answer: A
Correct Answer
Answer A is correct because CSV-to-labeled-object transformation directly fits. It can map each delimited column to its header name before inserting the record into model input. The scenario is specifically about the case where raw comma-separated values are ambiguous without field labels.
Incorrect Answers
Answer B is incorrect because ordered conversation-turn serialization targets a neighboring concern. It can preserve chronological role order when converting stored chat history into the model request when reordered dialog turns would change the meaning of follow-up questions. The key requirement here is raw comma-separated values are ambiguous without field labels.
Answer C is incorrect because speaker-label mapping to conversation roles is designed to map source labels such as AGENT and CUSTOMER into the model API’s supported conversational roles. That makes sense when source-system speaker names are not directly valid message roles for the inference API. The question is centered on raw comma-separated values are ambiguous without field labels.
Answer D is incorrect because JSON escaping of untrusted text fields can use a serializer that escapes quotes, backslashes, and control characters instead of string concatenation. Its best fit is where raw user text can otherwise produce malformed JSON request bodies. The current case instead needs raw comma-separated values are ambiguous without field labels.
Question 14
An application team is standardizing data before foundation-model inference. A platform team needs to add models without scattering provider-specific field names throughout business code. Which solution most directly resolves this issue? Choose ONE.
- normalized UTF-8 text encoding
- conversation attachment content blocks
- CSV-to-labeled-object transformation
- provider-versioned request adapter
Correct Answer: D
Correct Answer
Answer D is correct because provider-versioned request adapter should be used. It can route each selected model through an adapter that owns that model family’s request and response mapping. That capability is needed because an application supports multiple model families whose native schemas evolve independently.
Incorrect Answers
Answer A is incorrect because normalized UTF-8 text encoding can help a pipeline convert incoming text to a consistent Unicode encoding before request serialization, especially when mixed encodings create invalid characters and inconsistent tokenization. The present requirement is an application supports multiple model families whose native schemas evolve independently, so it is not the best answer.
Answer B is incorrect because conversation attachment content blocks would improve another stage by enabling it to represent text and supported attachments as separate typed content blocks within one conversational message. It fits when a message combines written instructions with a document or image and each part must retain its type. The stated problem is an application supports multiple model families whose native schemas evolve independently.
Answer C is incorrect because CSV-to-labeled-object transformation works for cases where raw comma-separated values are ambiguous without field labels by helping to map each delimited column to its header name before inserting the record into model input. The present scenario needs an application supports multiple model families whose native schemas evolve independently.
Question 15
A platform team is reviewing a production GenAI data pipeline. A multilingual feed combines UTF-8 and legacy Windows-1252 text, causing malformed characters in otherwise valid JSON requests. Which solution most directly resolves this issue? Choose ONE.
- model-specific Bedrock request serialization
- base64 or binary image content with correct media type
- JSON escaping of untrusted text fields
- normalized UTF-8 text encoding
Correct Answer: D
Correct Answer
Answer D is correct because normalized UTF-8 text encoding matches the constraint. It can convert incoming text to a consistent Unicode encoding before request serialization. This is the intended approach when mixed encodings create invalid characters and inconsistent tokenization.
Incorrect Answers
Answer A is incorrect because Using model-specific Bedrock request serialization would let the team build the JSON body expected by the selected model instead of assuming every provider uses the same native schema. That is appropriate when native InvokeModel payloads differ among model providers. The scenario instead calls for mixed encodings create invalid characters and inconsistent tokenization.
Answer B is incorrect because base64 or binary image content with correct media type has value because it can encode image content using the request format and MIME type expected by the multimodal API. Its proper fit is when visual input is valid but the request currently treats binary bytes as ordinary text. This question requires mixed encodings create invalid characters and inconsistent tokenization.
Answer C is incorrect because JSON escaping of untrusted text fields can help a pipeline use a serializer that escapes quotes, backslashes, and control characters instead of string concatenation, especially when raw user text can otherwise produce malformed JSON request bodies. The present requirement is mixed encodings create invalid characters and inconsistent tokenization, so it is not the best answer.
Question 16
A cloud engineering team must correct a data-processing weakness. Two customer names look identical on screen but compare differently because one source uses composed characters and another uses combining marks. Which solution most directly resolves this issue? Choose ONE.
- Unicode normalization
- canonical casing for identifiers
- navigation-boilerplate removal
- exact-content deduplication
Correct Answer: A
Correct Answer
Answer A is correct because Unicode normalization is correct. It convert canonically equivalent Unicode sequences to one representation before deduplication and inference. This matters because visually identical text is represented by different code-point sequences. That directly addresses the scenario.
Incorrect Answers
Answer B is incorrect because canonical casing for identifiers addresses another concern by helping to normalize case only for identifiers defined as case-insensitive while preserving case-sensitive free text. It fits where the same identifier appears in multiple letter cases and should match consistently. Here the deciding need is visually identical text is represented by different code-point sequences.
Answer C is incorrect because Using navigation-boilerplate removal would let the team strip repeated menus, cookie notices, and site chrome while retaining article content. That is appropriate when web-page boilerplate consumes context and distracts from the source information. The scenario instead calls for visually identical text is represented by different code-point sequences.
Answer D is incorrect because exact-content deduplication has value because it can remove repeated identical text blocks before assembling the model context. Its proper fit is when duplicated passages waste tokens and can overweight one source. This question requires visually identical text is represented by different code-point sequences.
Question 17
A regulated workflow requires deterministic preprocessing before inference. A maintenance corpus mixes temperatures in Fahrenheit and Celsius. The FM must compare readings consistently while auditors can still see the original values. Which solution most directly resolves this issue? Choose ONE.
- unit normalization with original-value retention
- language detection and language-specific routing
- navigation-boilerplate removal
- Bedrock-assisted text reformatting with constrained output
Correct Answer: A
Correct Answer
Answer A is correct because unit normalization with original-value retention directly fits. It can convert measurements to a common unit while preserving the source value for traceability. The scenario is specifically about the case where mixed units make comparison unreliable unless values are standardized.
Incorrect Answers
Answer B is incorrect because language detection and language-specific routing can detect the dominant language and route the record to a language-appropriate prompt or supported model path. It fits when documents arrive in several languages and the pipeline must select processing that matches each language. This scenario instead requires mixed units make comparison unreliable unless values are standardized, so it targets a different problem.
Answer C is incorrect because navigation-boilerplate removal is useful when web-page boilerplate consumes context and distracts from the source information because it can strip repeated menus, cookie notices, and site chrome while retaining article content. The current requirement is mixed units make comparison unreliable unless values are standardized, which makes this a mismatch.
Answer D is incorrect because Bedrock-assisted text reformatting with constrained output addresses another concern by helping to use an FM preprocessing step to transform irregular prose into a defined structure while validating the result afterward. It fits where source text is semistructured and deterministic parsing alone cannot reliably produce the required representation. Here the deciding need is mixed units make comparison unreliable unless values are standardized.
Question 18
A developer is hardening an FM-consumption workflow. A pricing table has separate columns for region, tier, and discount. Plain text extraction currently loses which value belongs to which header. Which solution most directly resolves this issue? Choose ONE.
- speaker-turn cleanup and labeling
- Amazon Comprehend entity extraction
- language detection and language-specific routing
- table structure preservation
Correct Answer: D
Correct Answer
Answer D is correct because table structure preservation should be used. It can represent headers and row relationships explicitly instead of flattening a table into disconnected text. That capability is needed because the model needs relational structure to interpret cell values accurately.
Incorrect Answers
Answer A is incorrect because speaker-turn cleanup and labeling belongs in workflows where a transcript contains repeated filler markers and ambiguous speaker boundaries because it can remove transcription artifacts and preserve who said each utterance before passing a conversation to the FM. This question instead asks for the model needs relational structure to interpret cell values accurately.
Answer B is incorrect because Amazon Comprehend entity extraction is useful when explicit entity metadata will improve downstream context and routing and can identify entities such as organizations, people, and locations before enriching the model input. The described workload requires the model needs relational structure to interpret cell values accurately.
Answer C is incorrect because language detection and language-specific routing targets a different issue by helping to detect the dominant language and route the record to a language-appropriate prompt or supported model path. It fits where documents arrive in several languages and the pipeline must select processing that matches each language. The deciding condition is the model needs relational structure to interpret cell values accurately.
Question 19
An application team is standardizing data before foundation-model inference. Free-form technician notes must be converted into fields for symptom, action, and outcome before a later FM performs fleet analysis. Which solution most directly resolves this issue? Choose ONE.
- Amazon Comprehend entity extraction
- Bedrock-assisted text reformatting with constrained output
- low-quality record filtering
- speaker-turn cleanup and labeling
Correct Answer: B
Correct Answer
Answer B is correct because Bedrock-assisted text reformatting with constrained output matches the constraint. It can use an FM preprocessing step to transform irregular prose into a defined structure while validating the result afterward. This is the intended approach when source text is semistructured and deterministic parsing alone cannot reliably produce the required representation.
Incorrect Answers
Answer A is incorrect because Amazon Comprehend entity extraction would be sound if the goal were explicit entity metadata will improve downstream context and routing; it can identify entities such as organizations, people, and locations before enriching the model input. The requirement here is source text is semistructured and deterministic parsing alone cannot reliably produce the required representation.
Answer C is incorrect because low-quality record filtering belongs in workflows where poor source quality is known and can be scored before inference because it can exclude records that fail a measured quality threshold instead of allowing noisy inputs to degrade response consistency. This question instead asks for source text is semistructured and deterministic parsing alone cannot reliably produce the required representation.
Answer D is incorrect because speaker-turn cleanup and labeling is useful when a transcript contains repeated filler markers and ambiguous speaker boundaries and can remove transcription artifacts and preserve who said each utterance before passing a conversation to the FM. The described workload requires source text is semistructured and deterministic parsing alone cannot reliably produce the required representation.
Question 20
A platform team is reviewing a production GenAI data pipeline. Customer account codes such as acme-42 and ACME-42 refer to the same account, but narrative text must retain its original capitalization. Which solution most directly resolves this issue? Choose ONE.
- navigation-boilerplate removal
- canonical casing for identifiers
- low-quality record filtering
- Amazon Comprehend entity extraction
Correct Answer: B
Correct Answer
Answer B is correct because canonical casing for identifiers is correct. It normalize case only for identifiers defined as case-insensitive while preserving case-sensitive free text. This matters because the same identifier appears in multiple letter cases and should match consistently. That directly addresses the scenario.
Incorrect Answers
Answer A is incorrect because navigation-boilerplate removal solves a different input problem. It helps to strip repeated menus, cookie notices, and site chrome while retaining article content, normally when web-page boilerplate consumes context and distracts from the source information. The current workflow needs the same identifier appears in multiple letter cases and should match consistently.
Answer C is incorrect because low-quality record filtering would be sound if the goal were poor source quality is known and can be scored before inference; it can exclude records that fail a measured quality threshold instead of allowing noisy inputs to degrade response consistency. The requirement here is the same identifier appears in multiple letter cases and should match consistently.
Answer D is incorrect because Amazon Comprehend entity extraction can identify entities such as organizations, people, and locations before enriching the model input, which matters when explicit entity metadata will improve downstream context and routing. The present case is driven by the same identifier appears in multiple letter cases and should match consistently.
