The correct answer is Amazon Bedrock, AWS’s fully managed service for building and scaling generative AI applications using foundation models (FMs). Bedrock gives developers access to models from leading providers such as Anthropic (Claude), Meta (Llama), Mistral, Cohere, and Amazon Titan. Users can invoke these models via API without managing infrastructure or model training. According to AWS documentation, Bedrock supports tasks such as text generation, summarization, question answering, image generation, and RAG workflows with minimal setup. It supports both on-demand and provisioned throughput modes and integrates with features like Guardrails, Knowledge Bases, and Agents for secure, enterprise-grade applications. Amazon Q Developer is a generative AI tool for developers, but it doesn’t host or scale models. Amazon Kendra is an intelligent search engine, and Amazon Comprehend is used for NLP tasks like entity extraction—not foundation model access.

Referenced AWS AI/ML Documents and Study Guides:

Amazon Bedrock Developer Guide – Foundation Models and Use Cases

AWS Certified ML Specialty Guide – Generative AI on AWS

Providing examples of text passages with corresponding positive or negative labels in the prompt followed by the new text passage to be classified is the correct prompt engineering strategy for using a large language model (LLM) on Amazon Bedrock for sentiment analysis.

Example-Driven Prompts:

This strategy, known as few-shot learning, involves giving the model examples of input-output pairs (e.g., text passages with their sentiment labels) to help it understand the task context.

It allows the model to learn from these examples and apply the learned pattern to classify new text passages correctly.

Why Option A is Correct:

Guides the Model: Providing labeled examples teaches the model how to perform sentiment analysis effectively, increasing accuracy.

Contextual Relevance: Aligns the model ' s responses to the specific task of classifying sentiment.

Why Other Options are Incorrect:

B. Detailed explanation of sentiment analysis: Unnecessary for the model ' s operation; it requires examples, not explanations.

C. New text passage without context: Provides no guidance or learning context for the model.

D. Unrelated task examples: Would confuse the model and lead to inaccurate results.

The verified answer is C. Faithfulness. The company is evaluating foundation models for a summarization task by using an LLM-as-a-judge evaluation job in Amazon Bedrock. AWS documentation states that with a model evaluation job that uses a judge model, Amazon Bedrock uses one LLM to score another model’s responses and provide an explanation of how it scored each prompt and response pair. AWS also states that Amazon Bedrock provides built-in metrics that can be selected for these judge-based evaluation jobs.

Among the answer choices, Faithfulness is the only valid built-in metric for an Amazon Bedrock LLM-as-a-judge model evaluation job. AWS API documentation lists the valid built-in metric names for LLM-as-a-judge evaluation jobs, including Correctness, Completeness, Faithfulness, Helpfulness, Coherence, Relevance, FollowingInstructions, ProfessionalStyleAndTone, and responsible AI metrics such as Harmfulness, Stereotyping, and Refusal. The same documentation identifies Summarization as a valid task type for model evaluation jobs. Therefore, Faithfulness is a valid built-in metric for evaluating whether generated summaries remain grounded in the source input rather than introducing unsupported claims.

Option A. Context relevance and option B. Context coverage are incorrect for this question because AWS lists those as metrics for knowledge base retrieval-only evaluation jobs, not general LLM-as-a-judge model evaluation of foundation model summarization output. These metrics evaluate retrieved context, not the quality of a summarization model’s generated response. Option D. RMSE is also incorrect because root mean square error is a regression metric. It is not a built-in LLM-as-a-judge metric for summarization in Amazon Bedrock.

Because the task is summarization and the evaluation job is LLM-as-a-judge, the correct built-in evaluation metric from the available choices is Faithfulness.

AWS documentation explains that data masking is a technique used to obscure sensitive information while preserving the usability of the dataset for analytics and machine learning. When sharing data with third parties, AWS recommends masking only the fields that contain personally identifiable information (PII) to minimize data exposure while maintaining data utility.

Applying data masking to PII fields ensures that sensitive attributes such as names, email addresses, phone numbers, or identification numbers are replaced with anonymized or obfuscated values. This allows the third-party provider to train machine learning models without accessing real personal data.

Masking all fields would unnecessarily reduce the dataset’s usefulness and could negatively impact model performance. Data encryption protects data at rest or in transit, but once the third party decrypts the dataset for processing, PII would still be exposed. Data labeling helps identify sensitive fields but does not prevent exposure.

AWS emphasizes that selective masking is a best practice for privacy-preserving data sharing, particularly in ML workflows where data structure and statistical properties must remain intact. Therefore, applying data masking only to PII fields is the correct solution.

Inference is the correct answer because it is the AI process that occurs when a deployed model analyzes new data (such as an image) to make predictions or identify objects.

Inference:

In the context of machine learning, inference is the process of using a trained model to make predictions on new, unseen data.

When the deep learning model is deployed to production and receives a new image for analysis, it uses the learned patterns from the training phase to identify objects in the image. This is known as inference.

Why Option B is Correct:

Inference Process: Involves applying the trained model to real-world data (the new image) to identify objects.

Deployment Context: The model has already been trained, and the deployment to production indicates it is being used for inference.

Why Other Options are Incorrect:

A. Training: Refers to the process of teaching the model using historical data, not making predictions on new data.

C. Model deployment: Refers to the process of making a trained model available for use in production.

D. Bias correction: Is a process to adjust a model to minimize bias, not for analyzing new images.

Amazon Augmented AI (Amazon A2I) is designed to enable human review workflows for machine learning predictions where human judgment is required. AWS documentation states that A2I allows organizations to integrate human reviewers into AI workflows to review low-confidence predictions or samples selected by business rules. This directly aligns with the requirement to involve employees in validating and improving translation outputs.

In this scenario, the translation model produces a confidence score, which is a common trigger used by Amazon A2I to route predictions to human reviewers. AWS explicitly describes that A2I supports use cases such as natural language processing, translation, content moderation, and document processing, where automated models may need human oversight to ensure accuracy and quality.

Amazon A2I provides managed workflows, reviewer task interfaces, and audit trails that allow employees to review, correct, and validate model outputs. The feedback collected through A2I can then be used to improve future model training, increasing translation quality over time.

The other options do not meet the requirement. Amazon SageMaker Clarify focuses on bias detection and explainability, not human review workflows. Amazon SageMaker Model Monitor is used to detect data drift and model performance degradation in production, not to involve humans in validating predictions. Amazon Bedrock Agents are designed to orchestrate tasks and interactions with foundation models, not to manage human-in-the-loop review processes.

AWS positions Amazon A2I as a core service for implementing human-in-the-loop machine learning, making it the correct solution for incorporating employees into a structured translation review process.

Sampling bias is the correct type of bias affecting the model output when it disproportionately flags people from a specific ethnic group.

Sampling Bias:

Occurs when the training data is not representative of the broader population, leading to skewed model outputs.

In this case, if the model disproportionately flags people from a specific ethnic group, it likely indicates that the training data was not adequately balanced or representative.

Why Option B is Correct:

Reflects Data Imbalance: A biased sample in the training data could result in unfair outcomes, such as disproportionately flagging a particular group.

Common Issue in ML Models: Sampling bias is a known problem that can lead to unfair or inaccurate model predictions.

Why Other Options are Incorrect:

A. Measurement bias: Involves errors in data collection or measurement, not sampling.

C. Observer bias: Refers to bias introduced by researchers or data collectors, not the model ' s output.

D. Confirmation bias: Involves favoring information that confirms existing beliefs, not relevant to model output bias.

Foundation models have a context window (max tokens), which limits the size of the input text (prompt + instructions).

If the input (e.g., a very long book) exceeds this limit, the model cannot process it, causing failure.

Temperature (A) and Top P (C) control randomness, not input size.

Fine-tuning (B) is irrelevant to input truncation failures.

???? Reference:

AWS Documentation – Amazon Bedrock Model Parameters (context size limits)

The correct answer is D – F1 score, which is the harmonic mean of precision and recall. AWS documentation states that the F1 score is the preferred metric when evaluating ML models trained on imbalanced datasets, where one class appears significantly more often than others. Accuracy becomes misleading in these situations because a model can achieve high accuracy simply by predicting the majority class. Precision alone measures correctness of positive predictions, while recall measures completeness—neither fully captures class balancing performance. The F1 score provides a combined measurement that reflects how well the model identifies minority classes while avoiding excessive false positives. AWS’s ML Specialty Guide recommends using F1 for classification tasks where “class distribution is uneven or where both false negatives and false positives matter.” This metric provides a fairer assessment of overall model performance across all class categories.

Referenced AWS Documentation:

AWS Certified Machine Learning Specialty Guide – Metrics for Imbalanced Classes

Amazon SageMaker Developer Guide – Evaluating Classification Models

Generative AI models (such as large language models) are designed to generate new content, such as text, summaries, images, and more. Summarizing text—like customer complaints—is a classic application of generative AI.

A is correct:

" Text summarization is a core generative AI use case, as it involves generating new, concise content from a larger body of text. "

(Reference: AWS Generative AI Use Cases)

B and C are standard ML classification/segmentation tasks.

D is a regression/prediction task, not generative.