Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 6

Question 50

An invoice archive mixes USD, EUR and GBP amounts that are labeled inconsistently before semantic indexing. What normalization avoids misleading comparisons?

Strip all currency symbols and embed only the number as text.
Parse the numeric amount with its original ISO currency and unit, store explicit normalized fields, and apply approved conversion only with a dated rate.
Assume every non-dollar amount is automatically USD at a one-to-one conversion.
Raise the embedding dimension until it represents currency conversions exactly.

Explanation

Correct Answer

An amount without its currency is not an equivalent quantity. Keeping source denomination and an auditable conversion date avoids comparing unrelated numerical values in retrieval-generated totals.

Incorrect Answers

The same digits in different currencies have different value; dropping currency metadata destroys their meaning.

An arbitrary exchange assumption changes financial values without a source or effective date.

Embedding dimensionality does not provide deterministic FX conversion or attach missing monetary units.

Question 51

A payroll dataset contains employee salary columns that support users are not authorized to retrieve. The data will be embedded in a shared search store. What should happen before embedding?

Embed all salary columns and rely on a friendly system prompt never to mention pay.
Give every support employee access to the entire index but disable result citations.
Encrypt the index at rest but leave unrestricted query permissions to salary data.
Remove or mask restricted salary fields under the source-data policy and enforce document-level access filters on the indexed content.

Explanation

Correct Answer

Embedding can preserve information about source text; unapproved sensitive fields should be minimized before indexing, and searchable records remain subject to their effective access rights.

Incorrect Answers

A prompt is not an access-control boundary, and sensitive source values have already entered the shared retrieval substrate.

Withholding citations does not prevent private values from being retrieved or paraphrased.

Encryption protects storage, not authorization of an already authenticated user to read restricted content.

Question 52

An invoice intake mixes scanned image-only PDFs with selectable-text PDFs. The default text extractor produces empty results for the scans. Which preprocessing design addresses both?

Route image-only scans through a supported OCR or multimodal document parser while using text extraction for text PDFs, preserving source-page associations.
Treat every scanned PDF as text and index the empty extraction result.
Resize image scans until the default text-only parser recognizes them as ordinary text.
Generate a synthetic invoice summary without reading the scanned pixels.

Explanation

Correct Answer

A text-only parser ignores pixels; supported Bedrock advanced parsers or document OCR can recover visual content. Source-page linkage keeps recovered text traceable to the original invoice.

Incorrect Answers

A scan does not contain a selectable text layer, so the chosen extraction path loses the document.

Changing pixel resolution does not create a PDF text layer or add OCR to a text-only parser.

A model cannot provide reliable source-grounded amounts from a document it never parsed.

Question 53

A recorded-call archive must become searchable by the words actually spoken, and investigators must locate the relevant moment in the audio. Select TWO processing outputs needed before semantic indexing.

Embed an empty transcript and assume a high similarity score will recover the spoken words.
Remove all speaker and timing information to shorten the context regardless of the audit requirement.
Run supported speech-to-text processing on the actual audio and store the resulting transcript.
Use file upload timestamps as a substitute for where within the call a phrase was spoken.
Retain the transcript segment timestamps and available speaker/language metadata linked to the source recording.

Explanation

Correct Answer

Correct: Amazon Transcribe or supported Bedrock multimodal parsing can recover spoken text for a text search representation.

Correct: source-linked segment timing and available speaker/language context make the transcript auditable against the original audio.

Incorrect Answers

Without recognized speech content, a text embedding cannot index what the callers said.

Removing alignment information prevents investigators from locating and verifying a cited spoken segment.

An S3 modification time describes the file object, not the timestamp of an utterance within the recording.

Question 54

The spreadsheet parser returns invoice totals as display strings, including a subtotal and tax. The GenAI assistant must reconcile the numeric grand total. What preprocessing gate is required?

Parse monetary cells into numeric values with units and deterministically recompute the stated subtotal, tax and total before accepting the extraction.
Use embedding cosine similarity between the subtotal and total strings as a replacement for arithmetic.
Ask the model to invent a balancing difference whenever the worksheet total fails.
Require every monetary cell to be non-empty without validating its numeric type or relationship.

Explanation

Correct Answer

An extracted table can be syntactically readable while its numerical values or formulas are wrong. Typed arithmetic validation catches discrepancies that an FM summary must not silently correct by guesswork.

Incorrect Answers

Two semantically similar amounts need not add correctly, so vector similarity cannot prove a financial equation.

An inferred adjustment would conceal source errors rather than identify the failed arithmetic check.

Completeness alone permits malformed values and mathematically inconsistent totals.

Question 55

A financial-report PDF contains bar charts whose titles are extracted but the bars and legend values vanish in a text-only knowledge-base parser. What is the supported ingestion fix?

Increase the vector-index dimensions without changing the parser.
Use a supported visual-aware Bedrock parsing strategy to extract chart/table content and retain links to the original page or figure.
Re-embed the extracted chart title and treat it as the full financial dataset.
Assume the text-only parser emits all pixels as typed table cells.

Explanation

Correct Answer

The default text parser cannot recover information encoded only in charts. Bedrock Data Automation or supported foundation-model parsing can extract visual elements and maintain source context.

Incorrect Answers

Higher-dimensional embeddings cannot recover bar heights or legend values absent from the source extraction.

A title does not contain the plotted values and cannot support quantitative conclusions.

AWS distinguishes text-only parsing from advanced multimodal parsing; unsupported visual elements may be ignored.

Question 56

During document preprocessing, a regulation table is split into retrieval chunks; the exported normalized rows omit the original column headings and year labels. Which ingestion transformation preserves each cell's meaning before embeddings are created?

Summarize the whole document before recovering the table headers and row fields.
Increase vector dimension to recover the missing column names from numerical similarity.
Serialize each extracted table row with its explicit column names, associated year labels, and source-page identifier before chunk embedding.
Increase top-k while retaining row fragments with no header metadata.

Explanation

Correct Answer

Correct: representing row values as named fields before embedding preserves which year and column each amount belongs to. Without that mapping, the FM can misattribute the extracted values even if the row text is retrieved.

Incorrect Answers

Summarizing the entire PDF before parsing does not preserve the column-to-value relationships lost when individual table rows were flattened.

A larger embedding dimension does not reconstruct headers omitted from the normalized row representation.

Raising the number of retrieved chunks cannot deterministically recover header associations that the parser never stored.

Question 57

A German-language CSV uses comma decimal separators (for example 1,25) and semicolons as delimiters. The extraction workflow currently parses values as text. What should be fixed?

Remove every comma in a numeric field and assume 1,25 means 125.
Use a US-locale comma field delimiter with no explicit escape handling.
Ask the FM to guess which currency a malformed decimal number represented.
Configure locale-aware delimiter and decimal parsing and verify that normalized numeric values preserve the original quantities.

Explanation

Correct Answer

Locale is part of the input-data contract: blindly treating a comma as a field separator or deleting it can turn 1,25 into an incorrect number.

Incorrect Answers

That converts a decimal separator to an integer concatenation and changes its value.

The parser may split one decimal amount into separate columns rather than a single value.

Currency estimation is unrelated to proper locale parsing and cannot validate the numeric source.

Question 58

An upstream producer adds an unexpected nested object to a JSON feed that is contractually restricted to an older schema. The ingestion pipeline currently accepts it without inspection. What gate should be added?

Accept any new field so long as the overall JSON parses.
Validate each record against its declared schema version and reject or route unknown/incompatible fields before they enter the model context.
Use a higher maxTokens setting to absorb unknown nested objects.
Silently delete all nested objects, including required known fields, without reporting schema drift.

Explanation

Correct Answer

Schema-version validation detects structural drift. A deterministic route to remediation avoids silently giving the model fields the consuming workflow did not authorize or understand.

Incorrect Answers

Syntactic JSON validity does not establish compatibility with the versioned business schema.

More context capacity does not validate field names, types or contracts.

Indiscriminate truncation can lose valid required data and hide upstream contract regressions.

Question 59

A Bedrock Converse request is rejected because the application serializes role and content as an untyped flat string rather than the expected message blocks. What must the client change?

Send the same malformed string through ConverseStream and assume streaming relaxes the message schema.
Change the prompt temperature to force the service to reinterpret invalid request fields.
Build the Converse messages array with valid user/assistant roles and typed content blocks appropriate to the selected model and API.
Wrap the flat string in a vector embedding and omit role/content fields.

Explanation

Correct Answer

Converse accepts structured Message objects and ContentBlock variants; a flattened transcript lacking required fields is not an equivalent API request.

Incorrect Answers

Streaming changes response delivery, not required request-message structure.

The Bedrock API validates message shape before decoding, so sampling settings have no effect.

Embeddings are not Converse Message objects and do not satisfy the conversational request contract.

Question 60

A text-and-image request is rejected because the client sends `https://.../photo.png` as a plain text content field. The chosen Bedrock model supports images. Which request fix is required?

Move the image URL into the system prompt while keeping it plain text.
Choose a larger output token limit to make the current text block a valid image block.
Change the search index metric from dot product to cosine before the Converse call.
Supply the image through the model-supported Converse image content block and source format/bytes or supported location, respecting model and API limits.

Explanation

Correct Answer

Image reasoning requires a typed image payload, not a URL appearing in text. The exact model/API must support the image format and permitted source representation.

Incorrect Answers

Relocating a string does not convert it into image pixels for a model input.

Output length cannot alter ContentBlock structure or image input support.

Vector similarity settings do not serialize a missing image payload.

Pass Your Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Exam Easy!

Question 50

Explanation

Correct Answer

Incorrect Answers

Question 51

Explanation

Correct Answer

Incorrect Answers

Question 52

Explanation

Correct Answer

Incorrect Answers

Question 53

Explanation

Correct Answer

Incorrect Answers

Question 54

Explanation

Correct Answer

Incorrect Answers

Question 55

Explanation

Correct Answer

Incorrect Answers

Question 56

Explanation

Correct Answer

Incorrect Answers

Question 57

Explanation

Correct Answer

Incorrect Answers

Question 58

Explanation

Correct Answer

Incorrect Answers

Question 59

Explanation

Correct Answer

Incorrect Answers

Question 60

Explanation

Correct Answer

Incorrect Answers

AWS Certified Generative AI Developer - Professional AIP-C01 Practice Tests