AIF-C01 Practice Test 16 — 15 Questions

Question 1

A startup wants to quickly prototype a generative AI application for summarizing news articles. They have limited ML expertise and want minimal infrastructure management. Which AWS service should they use?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Amazon Bedrock with a foundation model accessed via API.. Amazon Bedrock is the correct choice because it provides pre-trained foundation models from leading AI providers via a simple API, requiring no ML expertise or infrastructure management. The startup can quickly prototype a summarization application by sending news articles to the API and receiving summaries without training or deploying models.

Answer

Amazon SageMaker to build and train a custom summarization model.

Answer

AWS Lambda with a custom Python script using the Hugging Face Transformers library.

Answer

Amazon EC2 instance running a pre-trained model from AWS Marketplace.

Question 2

A machine learning team is building a binary classifier using Amazon SageMaker. The dataset has 10,000 features and 1,000 samples. The model overfits severely. Which TWO approaches are MOST likely to reduce overfitting? (Choose two.)

Accepted Answer

Perform feature selection to reduce the number of features. Option C is correct because with 10,000 features but only 1,000 samples, the model has far more parameters than data points, so performing feature selection to reduce the number of features directly lowers model complexity and the variance that causes overfitting. Option D is correct because adding L2 regularization to the loss function penalizes large weights, shrinking them toward zero and constraining the model's effective capacity, which is a standard, effective remedy for overfitting. Options A, B, and E do not belong: increasing batch size to the full dataset mainly affects gradient noise and optimization dynamics rather than reducing model capacity, adding more layers increases model complexity and would worsen overfitting, and training for more epochs lets the model fit the training data even more closely, also worsening overfitting.

Answer

Increase the batch size to the full dataset

Answer

Use a neural network with more layers

Answer

Train the model for more epochs

Question 3

A developer is using Amazon Bedrock to generate responses from a foundation model and wants to receive the output as a stream to improve user experience. Which TWO statements about streaming responses are correct? (Select TWO.)

Accepted Answer

Streaming reduces perceived latency by displaying partial results as they are generated. Option A is correct because streaming improves perceived latency by delivering tokens incrementally as the foundation model generates them, so the user sees partial output immediately rather than waiting for the full completion. Option B is correct because Amazon Bedrock's Converse API supports streaming via the ConverseStream operation, which returns a stream of events containing incremental text deltas and metadata. Option C is incorrect because returning the entire response at once describes a non-streaming (synchronous) invocation, not streaming. Option D is incorrect because the InvokeModel API is a non-streaming operation; streaming with model-specific payloads is done through InvokeModelWithResponseStream. Option E is incorrect because inference parameters such as temperature and max_tokens can still be supplied in the request body when streaming.

Answer

Streaming returns the entire response at once

Answer

Streaming is supported by the InvokeModel API

Answer

Streaming cannot be used with temperature or max_tokens parameters

Question 4

A team is planning to build an ML model that recommends products to users based on their purchase history. Which AWS service is MOST suitable?

Accepted Answer

Amazon Personalize. Amazon Personalize is the correct choice because it is a fully managed service specifically designed to build real-time recommendation systems using user-item interaction data, such as purchase history. It uses deep learning algorithms like HRNN (Hierarchical Recurrent Neural Network) to personalize product recommendations for each user, making it the most suitable for this use case.

Answer

Amazon SageMaker

Answer

Amazon Forecast

Answer

Amazon Rekognition

Question 5

A company uses Amazon Bedrock and needs to log all model invocations for audit purposes. The logs must be stored in a central S3 bucket and also sent to CloudWatch Logs for real-time monitoring. Which configuration should they use?

Accepted Answer

Configure Amazon Bedrock model invocation logging to send logs to both S3 and CloudWatch Logs. Bedrock model invocation logging can be configured to send logs to both S3 and CloudWatch Logs simultaneously. This is a built-in feature of Bedrock logging settings.

Answer

Enable Amazon Macie to monitor Bedrock responses

Answer

Enable AWS CloudTrail to capture Bedrock API calls and configure CloudTrail logs to S3 and CloudWatch

Answer

Use AWS Lambda to capture responses and write to S3 and CloudWatch

Question 6

An ML team notices that the training accuracy is 99% but validation accuracy is only 72%. Which concept best describes this situation?

Accepted Answer

Overfitting. The model achieves 99% accuracy on training data but only 72% on validation data, which is a classic symptom of overfitting. Overfitting occurs when the model learns noise and specific patterns in the training set too well, failing to generalize to unseen data. This is often caused by excessive model complexity, such as too many layers in a neural network or too deep a decision tree, relative to the amount of training data.

Answer

Cross-validation error

Answer

Bias-variance tradeoff

Answer

Underfitting

Question 7

A company wants to use a rules-based approach to approve loan applications but finds that it cannot keep up with changing regulations. They have historical data with decisions. Which approach should they adopt?

Accepted Answer

Train a supervised machine learning model on historical decisions. Supervised machine learning can automatically learn patterns from historical loan decision data, adapting to changing regulations without manual rule updates. By training a model on labeled examples of approved and rejected applications, the system can generalize to new cases and adjust as the underlying regulatory logic shifts over time, which is precisely the limitation of a static rules-based approach.

Answer

Continue refining the rules-based system manually

Answer

Implement a deep neural network without feature engineering

Answer

Use unsupervised learning to cluster applications

Question 8

A company wants to build a customer support chatbot that answers questions based on a large internal knowledge base. Which AWS service is most suitable for implementing RAG to retrieve relevant documents?

Accepted Answer

Amazon Kendra. Amazon Kendra is an intelligent enterprise search service powered by machine learning that is specifically designed for retrieving relevant documents from large knowledge bases, making it ideal for RAG implementations. It supports natural language queries and returns precise answers with citations, which is exactly what a customer support chatbot needs to retrieve relevant documents.

Answer

Amazon Lex

Answer

Amazon Polly

Answer

Amazon Connect

Question 9

A developer is building a customer support chatbot using Amazon Bedrock Agents. The agent needs to retrieve order status by calling an external API. Which configuration enables the agent to call the API?

Accepted Answer

Define an action group with an API schema and Lambda function. Amazon Bedrock Agents use action groups to define custom actions that the agent can invoke, such as calling an external API. An action group requires an API schema (in OpenAPI 3.0 format) to describe the API operations and a Lambda function to execute the API call and return the result. This configuration enables the agent to dynamically retrieve order status by invoking the Lambda function based on the user's request.

Answer

Configure a knowledge base with the API documentation

Answer

Use a guardrail to allow API calls

Answer

Enable model caching for the API endpoint

Question 10

A data scientist is evaluating a logistic regression model for a binary classification task. The model's AUC-ROC score is 0.95 on the training set and 0.51 on the test set. What is the MOST likely issue?

Accepted Answer

The model is overfitting the training data. A large gap between training and test AUC-ROC indicates overfitting — the model memorizes training data but fails to generalize. Cross-validation can help detect and mitigate this. Data leakage or class imbalance could also contribute, but overfitting is the primary symptom.

Answer

The test set is too small

Answer

The learning rate is too low

Answer

The model is underfitting the training data

Question 11

A developer is using Amazon Bedrock with the Claude model for text summarization. The output sometimes includes inaccurate information. What is the best practice to reduce hallucinations?

Accepted Answer

Use retrieval augmented generation. Retrieval Augmented Generation (RAG) grounds the model's output in external, authoritative knowledge sources by retrieving relevant documents and injecting them into the prompt context. This directly reduces hallucinations because the model generates summaries based on factual retrieved data rather than relying solely on its parametric memory, which is the primary source of inaccuracies in text summarization tasks.

Answer

Use a larger model

Answer

Increase temperature

Answer

Decrease max tokens

Question 12

A team deployed a text generation model on Amazon Bedrock. They want to monitor for toxic content in model outputs. Which evaluation approach is MOST effective?

Accepted Answer

Use Amazon Bedrock Model Evaluation with toxicity metrics. Amazon Bedrock Model Evaluation with toxicity metrics is the most effective approach because it provides automated, built-in evaluation of model outputs for toxic content using predefined metrics, directly integrated with the Bedrock service. This eliminates the need for manual effort or custom filtering, ensuring consistent and scalable monitoring of harmful content.

Answer

Enable CloudWatch Logs and set a metric filter for toxic words

Answer

Use Amazon SageMaker Ground Truth for human annotation

Answer

Manually review a sample of outputs each week

Question 13

A company is implementing an AI governance framework for their machine learning models deployed on Amazon SageMaker. Which THREE actions should they include to manage the model lifecycle effectively? (Select THREE.)

Accepted Answer

Enable Amazon CloudWatch alarms for model accuracy metrics. Option A is correct because Amazon CloudWatch alarms on model accuracy metrics (e.g., via SageMaker Model Monitor's ModelQuality metrics) provide automated detection of drift or degradation, which is essential for governing model performance across the lifecycle. Option C is correct because a formal decommissioning policy ensures old models are retired in a controlled, auditable way, preventing stale or non-compliant models from continuing to serve predictions. Option E is correct because SageMaker Model Registry provides model versioning, approval workflows, and metadata tracking, which are foundational for managing the model lifecycle and governance. Option B is not appropriate because automatically deleting all previous model versions destroys the audit trail and rollback capability that governance frameworks require. Option D is not the right tool because CloudTrail records AWS API activity (control-plane events), not individual model inference requests; inference logging is handled by SageMaker endpoint data capture or CloudWatch Logs.

Answer

Automatically delete all previous model versions after deployment

Answer

Use AWS CloudTrail to track model inference requests

Question 14

A team is using Amazon SageMaker to train a deep learning model. The training job is taking too long. Which action is MOST likely to reduce training time while maintaining model quality?

Accepted Answer

Increase the number of training instances for distributed training. Increasing the number of training instances for distributed training (Option D) directly reduces training time by parallelizing the workload across multiple machines, which is the most effective approach for deep learning models that are computationally intensive. SageMaker's distributed training libraries (e.g., SageMaker Distributed Data Parallel) split the data and model across instances, enabling linear or near-linear speedups while preserving model quality through synchronized gradient updates.

Answer

Use incremental training

Answer

Enable managed spot training

Answer

Switch to a smaller instance type

Question 15

Which prompt engineering technique involves providing the model with a few examples of desired input-output pairs before asking it to complete a new instance?

Accepted Answer

Few-shot prompting. Few-shot prompting (Option D) is the correct technique because it explicitly involves providing the model with a few examples of desired input-output pairs before asking it to complete a new instance. This helps the model understand the pattern, format, or task from the examples, improving performance on tasks like classification, translation, or formatting without requiring fine-tuning.

Answer

Zero-shot prompting

Answer

Chain-of-thought prompting

Answer

System prompting