100% Latest & Updated Google Professional Machine Learning Engineer Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 70
An ML engineering team must resolve this production problem: A support assistant repeats a very large fixed policy prefix in every request, driving unnecessary processing time and tokens. Which action best addresses the actual technical requirement?
- Evaluate several compatible first-party and open or partner models in Model Garden.
- Use supported caching for stable repeated prompt context with appropriate invalidation.
- Select an available Imagen image-generation model from Model Garden with controlled prompts.
- Use an appropriate pretrained Document AI invoice processor and confidence-based review workflow.
Explanation
Correct Answer
B:Use supported caching for stable repeated prompt context with appropriate invalidation. Measure cache hit rate, p95 response time and billed token categories. Frequently changing context or unsupported models may not receive savings.
Incorrect Answers
A:That measure applies when an engineering group needs a small low-latency text model but accidentally benchmarks only the largest expensive foundation model. The required intervention here is different: Use supported caching for stable repeated prompt context with appropriate invalidation. The need is a support assistant repeats a very large fixed policy prefix in every request, driving unnecessary processing time and tokens.
C:That measure applies when a marketing team needs concept illustrations described in text and has no base photos to classify or enhance. The required intervention here is different: Use supported caching for stable repeated prompt context with appropriate invalidation. The need is a support assistant repeats a very large fixed policy prefix in every request, driving unnecessary processing time and tokens.
D:That measure applies when an accounts-payable service receives variable-layout scanned invoices and needs supplier names, totals and line items without training OCR from scratch. The required intervention here is different: Use supported caching for stable repeated prompt context with appropriate invalidation. The need is a support assistant repeats a very large fixed policy prefix in every request, driving unnecessary processing time and tokens.
Question 71
An engineer is choosing an appropriate design for this workload: A child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers. Which action best addresses the actual technical requirement?
- Use an appropriate available Veo generative-video workflow with reviewed prompts.
- Use an appropriate pretrained Document AI invoice processor and confidence-based review workflow.
- Apply supported safety settings and separate content-policy screening and human escalation.
- Select an available Imagen image-generation model from Model Garden with controlled prompts.
Explanation
Correct Answer
C:Apply supported safety settings and separate content-policy screening and human escalation. Track unsafe response rate and false blocks over a representative red-team suite. No safety filter eliminates every harmful answer or prompt-injection attack.
Incorrect Answers
A:That measure applies when an advertising prototype needs brief new moving footage from a storyboard, not a caption describing an uploaded video. The required intervention here is different: Apply supported safety settings and separate content-policy screening and human escalation. The need is a child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers.
B:That measure applies when an accounts-payable service receives variable-layout scanned invoices and needs supplier names, totals and line items without training OCR from scratch. The required intervention here is different: Apply supported safety settings and separate content-policy screening and human escalation. The need is a child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers.
D:That measure applies when a marketing team needs concept illustrations described in text and has no base photos to classify or enhance. The required intervention here is different: Apply supported safety settings and separate content-policy screening and human escalation. The need is a child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers.
Question 72
An ML platform team is resolving 3 independent workstreams. Workstream A: An accounts-payable service receives variable-layout scanned invoices and needs supplier names, totals and line items without training OCR from scratch. Workstream B: A call center wants searchable transcripts of long recorded calls in supported languages without custom acoustic model training. Workstream C: An advertising prototype needs brief new moving footage from a storyboard, not a caption describing an uploaded video. Select exactly THREE diagnostic observations, one directly addressing each separate workstream.
- Verify whether a supported Veo version and generation mode are available in the tenant and location.
- Inspect channel count, audio encoding, consent and supported recognition configuration.
- Compare supported catalog models by latency, context size, quality, cost and regional availability.
- Inspect an invoice sample for layout diversity, scan quality and structured field accuracy.
- Run adversarial prompt tests and inspect current model-specific safety control options.
- Check multimodal model input support, privacy constraints, and whether images are relevant.
Explanation
Correct Answers
A:The evidence for this case shows situation: Verify whether a supported Veo version and generation mode are available in the tenant and location. It directly examines the reported situation: An advertising prototype needs brief new moving footage from a storyboard, not a caption describing an uploaded video. The corresponding corrective action is: Use an appropriate available Veo generative-video workflow with reviewed prompts.
B:The evidence for this case shows situation: Inspect channel count, audio encoding, consent and supported recognition configuration. It directly examines the reported situation: A call center wants searchable transcripts of long recorded calls in supported languages without custom acoustic model training. The corresponding corrective action is: Use Cloud Speech-to-Text with an appropriate long-audio transcription workflow.
D:The evidence for this case shows situation: Inspect an invoice sample for layout diversity, scan quality and structured field accuracy. It directly examines the reported situation: An accounts-payable service receives variable-layout scanned invoices and needs supplier names, totals and line items without training OCR from scratch. The corresponding corrective action is: Use an appropriate pretrained Document AI invoice processor and confidence-based review workflow.
Incorrect Answers
C:Comparing supported catalog models by latency, context size, quality, cost and regional availability is relevant when an engineering group needs a small low-latency text model but accidentally benchmarks only the largest expensive foundation model; it does not establish the result required by this workstream.
E:Running adversarial prompt tests and inspect current model-specific safety control options is relevant when a child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers; it does not establish the result required by this workstream.
F:Checking multimodal model input support, privacy constraints, and whether images are relevant is relevant when an insurance claims tool needs a textual summary that combines a customer message with an attached claim photo; it does not establish the result required by this workstream.
Question 73
An ML platform team is resolving 3 independent workstreams. Workstream A: A retailer needs common object labels for new catalog photos without an internal visual classifier or labeled product corpus. Workstream B: An insurance claims tool needs a textual summary that combines a customer message with an attached claim photo. Workstream C: An engineering group needs a small low-latency text model but accidentally benchmarks only the largest expensive foundation model. Select exactly THREE diagnostic observations, one directly addressing each separate workstream.
- Inspect an invoice sample for layout diversity, scan quality and structured field accuracy.
- Check multimodal model input support, privacy constraints, and whether images are relevant.
- Compare supported catalog models by latency, context size, quality, cost and regional availability.
- Check sample image content and whether generic Vision API labels cover the required categories.
- Run adversarial prompt tests and inspect current model-specific safety control options.
- Inspect channel count, audio encoding, consent and supported recognition configuration.
Explanation
Correct Answers
B:The evidence for this case shows situation: Check multimodal model input support, privacy constraints, and whether images are relevant. It directly examines the reported situation: An insurance claims tool needs a textual summary that combines a customer message with an attached claim photo. The corresponding corrective action is: Use a supported Gemini multimodal model with a bounded evidence-grounded prompt.
C:The evidence for this case shows situation: Compare supported catalog models by latency, context size, quality, cost and regional availability. It directly examines the reported situation: An engineering group needs a small low-latency text model but accidentally benchmarks only the largest expensive foundation model. The corresponding corrective action is: Evaluate several compatible first-party and open or partner models in Model Garden.
D:The evidence for this case shows situation: Check sample image content and whether generic Vision API labels cover the required categories. It directly examines the reported situation: A retailer needs common object labels for new catalog photos without an internal visual classifier or labeled product corpus. The corresponding corrective action is: Use supported Cloud Vision API label detection on the images.
Incorrect Answers
A:Inspecting an invoice sample for layout diversity, scan quality and structured field accuracy is relevant when an accounts-payable service receives variable-layout scanned invoices and needs supplier names, totals and line items without training OCR from scratch; it does not establish the result required by this workstream.
E:Running adversarial prompt tests and inspect current model-specific safety control options is relevant when a child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers; it does not establish the result required by this workstream.
F:Inspecting channel count, audio encoding, consent and supported recognition configuration is relevant when a call center wants searchable transcripts of long recorded calls in supported languages without custom acoustic model training; it does not establish the result required by this workstream.
Question 74
Before changing the system, the team investigates the following behavior: A travel website must translate brief help articles into French and Japanese and has no aligned text corpus for training a bespoke translation model. Which observation most directly identifies the cause or requirements before a change?
- Review supported language pairs and critical brand vocabulary.
- Check sample image content and whether generic Vision API labels cover the required categories.
- Inspect channel count, audio encoding, consent and supported recognition configuration.
- Run adversarial prompt tests and inspect current model-specific safety control options.
Explanation
Correct Answer
A:The evidence for this case shows situation: Review supported language pairs and critical brand vocabulary. It directly examines the reported situation: A travel website must translate brief help articles into French and Japanese and has no aligned text corpus for training a bespoke translation model. The corresponding corrective action is: Use Cloud Translation API with approved terminology handling where available.
Incorrect Answers
B:This evidence concerns a separate issue: A retailer needs common object labels for new catalog photos without an internal visual classifier or labeled product corpus. It does not establish what is happening in the present case: A travel website must translate brief help articles into French and Japanese and has no aligned text corpus for training a bespoke translation model. The diagnostic evidence to examine here is Review supported language pairs and critical brand vocabulary.
C:This evidence concerns a separate issue: A call center wants searchable transcripts of long recorded calls in supported languages without custom acoustic model training. It does not establish what is happening in the present case: A travel website must translate brief help articles into French and Japanese and has no aligned text corpus for training a bespoke translation model. The diagnostic evidence to examine here is Review supported language pairs and critical brand vocabulary.
D:This evidence concerns a separate issue: A child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers. It does not establish what is happening in the present case: A travel website must translate brief help articles into French and Japanese and has no aligned text corpus for training a bespoke translation model. The diagnostic evidence to examine here is Review supported language pairs and critical brand vocabulary.
Question 75
An ML platform team is resolving 3 independent workstreams. Workstream A: A call center wants searchable transcripts of long recorded calls in supported languages without custom acoustic model training. Workstream B: An advertising prototype needs brief new moving footage from a storyboard, not a caption describing an uploaded video. Workstream C: A support assistant repeats a very large fixed policy prefix in every request, driving unnecessary processing time and tokens. Select exactly THREE diagnostic observations, one directly addressing each separate workstream.
- Inspect channel count, audio encoding, consent and supported recognition configuration.
- Run adversarial prompt tests and inspect current model-specific safety control options.
- Verify whether a supported Veo version and generation mode are available in the tenant and location.
- Inspect whether the selected managed model supports context or prompt caching and its pricing.
- Check multimodal model input support, privacy constraints, and whether images are relevant.
- Compare supported catalog models by latency, context size, quality, cost and regional availability.
Explanation
Correct Answers
A:The evidence for this case shows situation: Inspect channel count, audio encoding, consent and supported recognition configuration. It directly examines the reported situation: A call center wants searchable transcripts of long recorded calls in supported languages without custom acoustic model training. The corresponding corrective action is: Use Cloud Speech-to-Text with an appropriate long-audio transcription workflow.
C:The evidence for this case shows situation: Verify whether a supported Veo version and generation mode are available in the tenant and location. It directly examines the reported situation: An advertising prototype needs brief new moving footage from a storyboard, not a caption describing an uploaded video. The corresponding corrective action is: Use an appropriate available Veo generative-video workflow with reviewed prompts.
D:The evidence for this case shows situation: Inspect whether the selected managed model supports context or prompt caching and its pricing. It directly examines the reported situation: A support assistant repeats a very large fixed policy prefix in every request, driving unnecessary processing time and tokens. The corresponding corrective action is: Use supported caching for stable repeated prompt context with appropriate invalidation.
Incorrect Answers
B:Running adversarial prompt tests and inspect current model-specific safety control options is relevant when a child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers; it does not establish the result required by this workstream.
E:Checking multimodal model input support, privacy constraints, and whether images are relevant is relevant when an insurance claims tool needs a textual summary that combines a customer message with an attached claim photo; it does not establish the result required by this workstream.
F:Comparing supported catalog models by latency, context size, quality, cost and regional availability is relevant when an engineering group needs a small low-latency text model but accidentally benchmarks only the largest expensive foundation model; it does not establish the result required by this workstream.
Question 76
A platform team needs diagnostic evidence for this condition: An insurance claims tool needs a textual summary that combines a customer message with an attached claim photo. Which observation most directly identifies the cause or requirements before a change?
- Inspect an invoice sample for layout diversity, scan quality and structured field accuracy.
- Review supported language pairs and critical brand vocabulary.
- Check multimodal model input support, privacy constraints, and whether images are relevant.
- Run adversarial prompt tests and inspect current model-specific safety control options.
Explanation
Correct Answer
C:The evidence for this case shows situation: Check multimodal model input support, privacy constraints, and whether images are relevant. It directly examines the reported situation: An insurance claims tool needs a textual summary that combines a customer message with an attached claim photo. The corresponding corrective action is: Use a supported Gemini multimodal model with a bounded evidence-grounded prompt.
Incorrect Answers
A:This evidence concerns a separate issue: An accounts-payable service receives variable-layout scanned invoices and needs supplier names, totals and line items without training OCR from scratch. It does not establish what is happening in the present case: An insurance claims tool needs a textual summary that combines a customer message with an attached claim photo. The diagnostic evidence to examine here is Check multimodal model input support, privacy constraints, and whether images are relevant.
B:This evidence concerns a separate issue: A travel website must translate brief help articles into French and Japanese and has no aligned text corpus for training a bespoke translation model. It does not establish what is happening in the present case: An insurance claims tool needs a textual summary that combines a customer message with an attached claim photo. The diagnostic evidence to examine here is Check multimodal model input support, privacy constraints, and whether images are relevant.
D:This evidence concerns a separate issue: A child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers. It does not establish what is happening in the present case: An insurance claims tool needs a textual summary that combines a customer message with an attached claim photo. The diagnostic evidence to examine here is Check multimodal model input support, privacy constraints, and whether images are relevant.
Question 77
An ML platform team is resolving 3 independent workstreams. Workstream A: A marketing team needs concept illustrations described in text and has no base photos to classify or enhance. Workstream B: A company chatbot answers questions about frequently changing internal procedures but a pretrained LLM invents obsolete policy details. Workstream C: An accounts-payable service receives variable-layout scanned invoices and needs supplier names, totals and line items without training OCR from scratch. Select exactly THREE diagnostic observations, one directly addressing each separate workstream.
- Run adversarial prompt tests and inspect current model-specific safety control options.
- Inspect an invoice sample for layout diversity, scan quality and structured field accuracy.
- Measure unsupported answers and verify the internal corpus contains current approved documents.
- Check whether a supported Imagen model provides the requested generation mode and rights terms.
- Check multimodal model input support, privacy constraints, and whether images are relevant.
- Compare supported catalog models by latency, context size, quality, cost and regional availability.
Explanation
Correct Answers
B:The evidence for this case shows situation: Inspect an invoice sample for layout diversity, scan quality and structured field accuracy. It directly examines the reported situation: An accounts-payable service receives variable-layout scanned invoices and needs supplier names, totals and line items without training OCR from scratch. The corresponding corrective action is: Use an appropriate pretrained Document AI invoice processor and confidence-based review workflow.
C:The evidence for this case shows situation: Measure unsupported answers and verify the internal corpus contains current approved documents. It directly examines the reported situation: A company chatbot answers questions about frequently changing internal procedures but a pretrained LLM invents obsolete policy details. The corresponding corrective action is: Build retrieval-grounded generation with access-controlled search and source citations.
D:The evidence for this case shows situation: Check whether a supported Imagen model provides the requested generation mode and rights terms. It directly examines the reported situation: A marketing team needs concept illustrations described in text and has no base photos to classify or enhance. The corresponding corrective action is: Select an available Imagen image-generation model from Model Garden with controlled prompts.
Incorrect Answers
A:Running adversarial prompt tests and inspect current model-specific safety control options is relevant when a child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers; it does not establish the result required by this workstream.
E:Checking multimodal model input support, privacy constraints, and whether images are relevant is relevant when an insurance claims tool needs a textual summary that combines a customer message with an attached claim photo; it does not establish the result required by this workstream.
F:Comparing supported catalog models by latency, context size, quality, cost and regional availability is relevant when an engineering group needs a small low-latency text model but accidentally benchmarks only the largest expensive foundation model; it does not establish the result required by this workstream.
Question 78
An incident review is assessing this observed failure: An advertising prototype needs brief new moving footage from a storyboard, not a caption describing an uploaded video. Which observation most directly identifies the cause or requirements before a change?
- Inspect an invoice sample for layout diversity, scan quality and structured field accuracy.
- Verify whether a supported Veo version and generation mode are available in the tenant and location.
- Inspect channel count, audio encoding, consent and supported recognition configuration.
- Check multimodal model input support, privacy constraints, and whether images are relevant.
Explanation
Correct Answer
B:The evidence for this case shows situation: Verify whether a supported Veo version and generation mode are available in the tenant and location. It directly examines the reported situation: An advertising prototype needs brief new moving footage from a storyboard, not a caption describing an uploaded video. The corresponding corrective action is: Use an appropriate available Veo generative-video workflow with reviewed prompts.
Incorrect Answers
A:This evidence concerns a separate issue: An accounts-payable service receives variable-layout scanned invoices and needs supplier names, totals and line items without training OCR from scratch. It does not establish what is happening in the present case: An advertising prototype needs brief new moving footage from a storyboard, not a caption describing an uploaded video. The diagnostic evidence to examine here is Verify whether a supported Veo version and generation mode are available in the tenant and location.
C:This evidence concerns a separate issue: A call center wants searchable transcripts of long recorded calls in supported languages without custom acoustic model training. It does not establish what is happening in the present case: An advertising prototype needs brief new moving footage from a storyboard, not a caption describing an uploaded video. The diagnostic evidence to examine here is Verify whether a supported Veo version and generation mode are available in the tenant and location.
D:This evidence concerns a separate issue: An insurance claims tool needs a textual summary that combines a customer message with an attached claim photo. It does not establish what is happening in the present case: An advertising prototype needs brief new moving footage from a storyboard, not a caption describing an uploaded video. The diagnostic evidence to examine here is Verify whether a supported Veo version and generation mode are available in the tenant and location.
Question 79
An ML platform team is resolving 3 independent workstreams. Workstream A: An engineering group needs a small low-latency text model but accidentally benchmarks only the largest expensive foundation model. Workstream B: A child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers. Workstream C: A travel website must translate brief help articles into French and Japanese and has no aligned text corpus for training a bespoke translation model. Select exactly THREE diagnostic observations, one directly addressing each separate workstream.
- Compare supported catalog models by latency, context size, quality, cost and regional availability.
- Run adversarial prompt tests and inspect current model-specific safety control options.
- Check sample image content and whether generic Vision API labels cover the required categories.
- Check multimodal model input support, privacy constraints, and whether images are relevant.
- Inspect channel count, audio encoding, consent and supported recognition configuration.
- Review supported language pairs and critical brand vocabulary.
Explanation
Correct Answers
A:The evidence for this case shows situation: Compare supported catalog models by latency, context size, quality, cost and regional availability. It directly examines the reported situation: An engineering group needs a small low-latency text model but accidentally benchmarks only the largest expensive foundation model. The corresponding corrective action is: Evaluate several compatible first-party and open or partner models in Model Garden.
B:The evidence for this case shows situation: Run adversarial prompt tests and inspect current model-specific safety control options. It directly examines the reported situation: A child-focused learning bot generates some unsafe text despite a system instruction asking for safe answers. The corresponding corrective action is: Apply supported safety settings and separate content-policy screening and human escalation.
F:The evidence for this case shows situation: Review supported language pairs and critical brand vocabulary. It directly examines the reported situation: A travel website must translate brief help articles into French and Japanese and has no aligned text corpus for training a bespoke translation model. The corresponding corrective action is: Use Cloud Translation API with approved terminology handling where available.
Incorrect Answers
C:Checking sample image content and whether generic Vision API labels cover the required categories is relevant when a retailer needs common object labels for new catalog photos without an internal visual classifier or labeled product corpus; it does not establish the result required by this workstream.
D:Checking multimodal model input support, privacy constraints, and whether images are relevant is relevant when an insurance claims tool needs a textual summary that combines a customer message with an attached claim photo; it does not establish the result required by this workstream.
E:Inspecting channel count, audio encoding, consent and supported recognition configuration is relevant when a call center wants searchable transcripts of long recorded calls in supported languages without custom acoustic model training; it does not establish the result required by this workstream.
Question 80
Before changing the system, the team investigates the following behavior: A company chatbot answers questions about frequently changing internal procedures but a pretrained LLM invents obsolete policy details. Which observation most directly identifies the cause or requirements before a change?
- Inspect whether the selected managed model supports context or prompt caching and its pricing.
- Inspect an invoice sample for layout diversity, scan quality and structured field accuracy.
- Inspect channel count, audio encoding, consent and supported recognition configuration.
- Measure unsupported answers and verify the internal corpus contains current approved documents.
Explanation
Correct Answer
D:The evidence for this case shows situation: Measure unsupported answers and verify the internal corpus contains current approved documents. It directly examines the reported situation: A company chatbot answers questions about frequently changing internal procedures but a pretrained LLM invents obsolete policy details. The corresponding corrective action is: Build retrieval-grounded generation with access-controlled search and source citations.
Incorrect Answers
A:This evidence concerns a separate issue: A support assistant repeats a very large fixed policy prefix in every request, driving unnecessary processing time and tokens. It does not establish what is happening in the present case: A company chatbot answers questions about frequently changing internal procedures but a pretrained LLM invents obsolete policy details. The diagnostic evidence to examine here is Measure unsupported answers and verify the internal corpus contains current approved documents.
B:This evidence concerns a separate issue: An accounts-payable service receives variable-layout scanned invoices and needs supplier names, totals and line items without training OCR from scratch. It does not establish what is happening in the present case: A company chatbot answers questions about frequently changing internal procedures but a pretrained LLM invents obsolete policy details. The diagnostic evidence to examine here is Measure unsupported answers and verify the internal corpus contains current approved documents.
C:This evidence concerns a separate issue: A call center wants searchable transcripts of long recorded calls in supported languages without custom acoustic model training. It does not establish what is happening in the present case: A company chatbot answers questions about frequently changing internal procedures but a pretrained LLM invents obsolete policy details. The diagnostic evidence to examine here is Measure unsupported answers and verify the internal corpus contains current approved documents.
Professional Machine Learning Engineer Practice Tests