Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 7

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.

Question 61

A generated tool request contains status="ARCHIVVED", but the backend API permits only OPEN, CLOSED and ARCHIVED. What prevents unauthorized action attempts?

Validate tool arguments against the typed schema and business enum before executing the requested backend operation.
Automatically add unrecognized enum values to the backend database and execute them.
Ask the model to restate the value in more natural language, then execute without schema checks.
Increase the context window while allowing any tool argument value.

Explanation

Correct Answer

A tool-call proposal is not a trusted command. Application or supported server-side execution must validate typed fields and permissions, returning an error for unsupported enum values.

Incorrect Answers

The model is not authorized to expand business-state definitions; this bypasses domain validation.

Text rephrasing is not a typed-value guarantee and does not constrain the actual API request.

Context capacity is unrelated to the tool API's permitted status domain.

Question 62

A Bedrock prompt occasionally exceeds the model's input budget after retrieval. The app reserves no space for its required final answer. What must the request assembler do?

Replace the input token count with the number of UTF-8 characters in the source files.
Set maxTokens to an unlimited value and submit all retrieved documents unchanged.
Count model-specific input tokens for supported requests, select relevant bounded evidence, and reserve response-token capacity before invoking the model.
Disable retrieval filtering and ask the model to skip unread text silently.

Explanation

Correct Answer

AWS CountTokens can estimate model-specific input size for supported model/endpoint combinations; the assembled prompt plus output reserve must respect model limits.

Incorrect Answers

Characters and model tokens are not equivalent, and request overhead/content-block encoding also matters.

A generation allowance cannot override a model context limit or accept unbounded retrieved evidence.

The oversized request can fail before generation; ignoring evidence after submission does not reduce the input token count.

Question 63

A legal pipeline normalizes Unicode and whitespace for retrieval, but an auditor later needs the exact original wording. Which provenance control protects the evidence?

Overwrite the original with only normalized text and rely on visual similarity.
Store only a vector embedding of the normalized text as the legal source of record.
Ask a foundation model to regenerate the original text later from memory.
Retain immutable raw source byte/text spans and a traceable normalized derivative with transformation metadata and offsets.

Explanation

Correct Answer

Unicode normalization can change code points or whitespace while preserving visual similarity. An auditable derivative must be linked to the original evidence for exact quotations and forensic review.

Incorrect Answers

A normalized string may not reproduce exact original bytes or distinguish transformation effects.

Embeddings are lossy representations and cannot reconstruct authoritative wording or code points.

Generative reconstruction is not reliable provenance and cannot prove a source was unaltered.

Question 64

OCR transcribes a printed serial number O0I1 as 0011. A model answer must reference the exact serial value on the scanned product label. What check is appropriate?

Use OCR confidence alone as proof every character matches the original image.
Drop all letters from serial numbers to simplify text matching.
Apply an expected serial-number pattern and reconcile low-confidence or ambiguous OCR characters against the original image before approval.
Increase the search-result count while retaining the wrong OCR serial string.

Explanation

Correct Answer

OCR can confuse visually similar glyphs; an application-specific pattern and verified source crop reduce false identifiers before retrieval or business lookup.

Incorrect Answers

A model can be confidently wrong on similar shapes; important serial values require rule and source checks.

Removing valid characters can create collisions and corrupt the identifier.

Retrieving more matches does not repair the misread source identifier.

Question 65

An entity extractor labels an employee named Jordan as an organization and incorrectly routes the record into a corporate directory. What preprocessing validation is necessary?

Apply entity-type classification with source context and validate high-risk person/company labels before using extracted identities.
Accept the first entity label produced by the model whenever a nonempty name exists.
Mask all proper nouns and then infer the entity type without their contextual evidence.
Choose a larger embedding store without assessing the entity label.

Explanation

Correct Answer

Names can be ambiguous; schema and context-aware entity checks distinguish PERSON from ORGANIZATION before identity-dependent workflows rely on model output.

Incorrect Answers

A valid name string does not establish whether it is a person or organization.

Removing distinguishing context can make classification less reliable.

Storage capacity does not validate the semantic type attached to a named entity.

Question 66

Users can upload arbitrary file types to a customer-support intake, but ingestion supports only a known set of safe document parsers. What boundary should the pipeline enforce?

Pass every uploaded binary to the first text parser and accept whatever string it emits.
Raise the retrieval token budget instead of validating file type.
Trust a user-supplied filename extension without inspecting the actual content type.
Validate allowed MIME types, sizes and parser routes before processing; reject unsupported content with safe error feedback.

Explanation

Correct Answer

Parsing untrusted uploads requires an allowlist and bounded resource limits. Rejecting unsupported input before invoking a parser avoids silently missing data and reduces malformed-content risk.

Incorrect Answers

Unknown MIME types may produce empty or misleading extraction and expose parsers to unsupported inputs.

Token budgets apply after extraction and cannot ensure a binary format is safe or supported.

Renaming a binary does not change its format, so extension-only acceptance is insufficient.

Question 67

An S3-based ingestion job drops document revision IDs during normalization. Two policy versions share the same object key. What metadata must survive the ETL process?

Keep only the friendly document title because it is stable across revisions.
Store the latest S3 modification time but discard object version identifiers.
Preserve the source bucket/key and, where S3 Versioning is enabled, its object version ID and effective-date lineage in normalized records.
Reconstruct the missing version ID from a semantic embedding when required.

Explanation

Correct Answer

S3 Versioning assigns immutable version IDs for distinct writes to one key. Preserving those references lets a normalized chunk be traced back to the exact source revision.

Incorrect Answers

A human title is not a unique S3 object version and cannot identify which text was indexed.

Timestamps alone are not a substitute for an explicit version identity when retrieving or auditing an exact historic object.

An embedding does not encode a retrievable S3 version ID or guarantee exact source lineage.

Question 68

A normalized evidence record must identify the precise versioned S3 object it came from and which ETL ingestion run produced it. Select THREE identifiers that must be carried through into its lineage metadata.

The original source bucket and object key.
The S3 object version ID for that source version.
Only the document display name, without the object key or version.
The ingestion run or job execution identifier.
The nearest-neighbor similarity score, without any source object reference.
The model temperature used later for answer generation.

Explanation

Correct Answer

Correct: bucket and key locate the logical S3 object but need a version identifier to distinguish historical writes.

Correct: S3 Versioning assigns an immutable version ID that identifies the exact object revision when enabled.

Correct: the ETL run ID identifies the processing execution responsible for creating this normalized record.

Incorrect Answers

A display title is not a stable bucket/key/version coordinate; multiple objects and revisions may share that title.

A similarity score describes retrieval ranking, not the identity of the source S3 object or processing run.

Decoding temperature belongs to downstream inference and cannot tell which document version an ETL job processed.

Question 69

A schema change causes a sudden spike in invalid records across a Glue batch; the ingestion run still reports success. Which operational response protects downstream GenAI context?

Suppress failure metrics and rely on a successful ETL exit code.
Index the damaged batch immediately and ask the FM to ignore suspicious records.
Replace invalid rows with arbitrary defaults and report a normal pass rate.
Publish quality-rule failure metrics, quarantine the affected batch and notify the responsible data owner before publishing new context.

Explanation

Correct Answer

Quality gates need observable outcomes and ownership. Isolating the faulty run prevents drifted source rows from being treated as validated production evidence.

Incorrect Answers

A process exit status does not prove row-level data quality; silent schema drift can propagate invalid context.

A prompt cannot enforce upstream quality acceptance or guarantee that invalid rows are not retrieved.

Invented defaults hide the regression and alter business records instead of preserving traceable failures.

Question 70

A Bedrock assistant request inadvertently includes internal debug headers containing employee email addresses and access tokens. The business question needs neither field. Which request-assembly control is required?

Send all debug fields and rely on the model to refrain from repeating them in the answer.
Add a larger output safety filter without changing the input fields.
Apply purpose-based allowlisting and remove unnecessary sensitive metadata before constructing the model input.
Move the raw debug headers into an unfiltered system message and keep them for future support.

Explanation

Correct Answer

Data minimization should happen before inference, logs or retrieval; irrelevant credentials and personal identifiers must not be forwarded as generative context.

Incorrect Answers

Once submitted, secrets are already in the inference payload regardless of response wording.

Output filtering does not eliminate unnecessary sensitive material in the request or its associated telemetry.

An alternate prompt role is not a safe storage or need-to-know boundary for secrets.

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

Question 60

Explanation

Correct Answer

Incorrect Answers

Question 61

Explanation

Correct Answer

Incorrect Answers

Question 62

Explanation

Correct Answer

Incorrect Answers

Question 63

Explanation

Correct Answer

Incorrect Answers

Question 64

Explanation

Correct Answer

Incorrect Answers

Question 65

Explanation

Correct Answer

Incorrect Answers

Question 66

Explanation

Correct Answer

Incorrect Answers

Question 67

Explanation

Correct Answer

Incorrect Answers

Question 68

Explanation

Correct Answer

Incorrect Answers

Question 69

Explanation

Correct Answer

Incorrect Answers

Question 70

Explanation

Correct Answer

Incorrect Answers

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