Question bank
AWS Certified AI Practitioner
41 AWS Certified AI Practitioner questions with the correct answer and a written explanation, organised across 5 exam domains.
- 41 questions
- 5 domains
- Foundation level
certilumQuestion index
Jump straight to any question. Answers and explanations are inline.
AI and ML Services on AWS
10 questions in this domain
- Question 1Single Choice
Which AWS service is used for automatic speech recognition to convert audio into text?
- AAmazon Polly
- BAmazon TranscribeCorrect answer
- CAmazon Rekognition
- DAmazon Comprehend
Why this is the answer
Amazon Transcribe is the AWS service for automatic speech recognition (ASR).
- Question 2Multiple Select
Which of the following are correct use cases for Amazon Rekognition? (Select two.)
- ADetecting objects in imagesCorrect answer
- BTranslating text between languages
- CAnalyzing sentiment in customer reviews
- DIdentifying faces in videosCorrect answer
Why this is the answer
Amazon Rekognition provides image and video analysis, including object detection and facial recognition.
- Question 3Interactive
Analyze the code block. What does this code do when invoked in a Python script using the AWS SDK (boto3)?
- ASynthesize speech from textCorrect answer
- BTranscribe audio to text
- CTranslate text to another language
- DDetect objects in an image
Why this is the answer
The code calls Amazon Polly's synthesize_speech method to convert text into an audio stream.
- Question 4Single Choice
Which SageMaker component is used to orchestrate the end-to-end ML workflow as a directed acyclic graph (DAG)?
- ASageMaker Studio
- BSageMaker PipelinesCorrect answer
- CSageMaker Model Registry
- DSageMaker Ground Truth
Why this is the answer
SageMaker Pipelines creates and manages ML workflows as DAGs.
- Question 5Multiple Select
Which two statements about Amazon SageMaker training jobs are correct? (Select two.)
- AThey can use managed spot instances to reduce cost.Correct answer
- BThey automatically deploy the model after training.
- CThey support distributed training across multiple compute instances.Correct answer
- DThey require the use of built-in algorithms only.
Why this is the answer
Training Jobs can use spot instances and distributed training; deployment is a separate step.
- Question 6Interactive
Examine the following SageMaker API call from the AWS CLI. What does the '--resource-type' value indicate about the resource being created?
- AA model version in Model Registry
- BA training job
- CA hyperparameter tuning jobCorrect answer
- DAn endpoint configuration
Why this is the answer
The command 'create-hyper-parameter-tuning-job' creates a tuning job to optimize hyperparameters.
- Question 7Single Choice
Which AWS service is designed to provide personalized product recommendations in real-time based on user behavior?
- AAmazon PersonalizeCorrect answer
- BAmazon Forecast
- CAmazon Rekognition
- DAmazon Comprehend
Why this is the answer
Amazon Personalize provides real-time personalization and recommendations.
- Question 8Multiple Select
Which two dataset types are required when creating a dataset group in Amazon Forecast? (Select two.)
- ATarget time seriesCorrect answer
- BRelated time series
- CItem metadata
- DTarget time series and item metadata are both required.Correct answer
Why this is the answer
A dataset group requires target time series and item metadata; related time series is optional.
- Question 9Interactive
The following Python code uses the AWS SDK to call a prediction service. Which service is being invoked?
- AAmazon PersonalizeCorrect answer
- BAmazon Forecast
- CAmazon SageMaker
- DAmazon Rekognition
Why this is the answer
The 'get_recommendations' method is part of the Amazon Personalize runtime API.
- Question 10Single Choice
Which AWS service helps automatically discover and classify sensitive data, such as PII, stored in Amazon S3?
- AAmazon GuardDuty
- BAmazon MacieCorrect answer
- CAWS WAF
- DAWS Shield
Why this is the answer
Amazon Macie uses ML to automatically discover, classify, and protect sensitive data in S3.
Data Preparation for ML
5 questions in this domain
- Question 11Single Choice
Which data type has a rigid, predefined schema and is stored in relational databases?
- AStructuredCorrect answer
- BUnstructured
- CSemi-structured
- DNone of the above
Why this is the answer
Structured data has a rigid, predefined schema and is stored in relational databases like Amazon RDS.
- Question 12Multiple Select
Which two of the following are data cleaning techniques mentioned in the document?
- ANormalization
- BDeduplicationCorrect answer
- CFeature engineering
- DHandling missing valuesCorrect answer
Why this is the answer
Data cleaning techniques include removing duplicate records and handling missing values. Normalization and feature engineering are separate preprocessing steps.
- Question 13Interactive
Given the following SQL query, which AWS service is being used to execute it?
- AAWS Glue
- BAmazon AthenaCorrect answer
- CAmazon Redshift
- DAmazon RDS
Why this is the answer
Amazon Athena is the serverless query service that runs SQL directly on data in S3.
- Question 14Single Choice
What is the minimum number of samples per category recommended for sufficiency in classification?
- A50
- B100Correct answer
- C500
- D1000
Why this is the answer
The document states that sufficiency requires at least 100 samples per category for classification.
- Question 15Multiple Select
Which two AWS services are specifically designed for processing unstructured data?
- AAmazon RekognitionCorrect answer
- BAmazon RDS
- CAmazon ComprehendCorrect answer
- DAWS Glue
Why this is the answer
Amazon Rekognition processes images and video, while Amazon Comprehend handles text; both are designed for unstructured data.
Deployment and Operations
8 questions in this domain
- Question 16Single Choice
What is the primary purpose of an Amazon SageMaker endpoint?
- AStore training data
- BHost trained models for inferenceCorrect answer
- CRun batch transform jobs
- DMonitor model drift
Why this is the answer
A SageMaker endpoint is a managed HTTPS service that hosts a trained model to serve real-time inference requests.
- Question 17Multiple Select
Which two drift types can Amazon SageMaker Model Monitor detect?
- AData driftCorrect answer
- BConcept driftCorrect answer
- CCode drift
- DInfrastructure drift
Why this is the answer
SageMaker Model Monitor detects data drift (changes in input feature distribution) and concept drift (changes in the relationship between features and target).
- Question 18Interactive
After executing the following AWS CLI command, what approval status is assigned to the registered model version?
- AApproved
- BPendingApprovalCorrect answer
- CRejected
- DAccepted
Why this is the answer
By default, when registering a model via the CLI without specifying an approval status, it is set to PendingApproval.
- Question 19Single Choice
What is a production variant in SageMaker?
- AA separate model container within an instance
- BA set of compute resources serving a model with a traffic weightCorrect answer
- CA version of the training dataset
- DAn auto-scaling policy
Why this is the answer
A production variant is a group of compute instances that serve a specific model and receive a weighted portion of inference traffic.
- Question 20Multiple Select
Which two components are required to set up SageMaker Model Monitor?
- AA baseline dataset with statisticsCorrect answer
- BData capture enabled on the endpointCorrect answer
- CAn Amazon EMR cluster
- DA SageMaker Notebook instance
Why this is the answer
Model Monitor requires a baseline (statistical summary of training data) and data capture on the endpoint to collect inference payloads.
- Question 21Interactive
The code snippet creates a SageMaker Pipeline step. What is the purpose of the 'ConditionStep' in this pipeline?
- AIt runs a training job in parallel
- BIt evaluates metrics and determines whether to register the modelCorrect answer
- CIt deploys the model to an endpoint
- DIt copies data from S3
Why this is the answer
The ConditionStep checks if evaluation metrics meet a threshold, and only if true, proceeds to register the model.
- Question 22Single Choice
What is the main advantage of using SageMaker Serverless Inference?
- AIt scales to zero when idle, reducing costCorrect answer
- BIt provides sub-millisecond latency always
- CIt supports longer timeouts than real-time endpoints
- DIt does not require any AWS service role
Why this is the answer
Serverless Inference automatically scales to zero during idle periods, so you pay only for invocations and compute duration.
- Question 23Multiple Select
Which two features are part of SageMaker Model Registry?
- AVersioning model artifactsCorrect answer
- BApproval workflow for model versionsCorrect answer
- CAutomatically training new models
- DReal-time monitoring of endpoints
Why this is the answer
Model Registry stores model versions and supports an approval lifecycle (PendingApproval, Approved, Rejected) to manage production deployments.
Fundamentals of AI and ML
8 questions in this domain
- Question 24Single Choice
Which machine learning paradigm uses labeled data to train a model to predict outcomes?
- ASupervised learningCorrect answer
- BUnsupervised learning
- CReinforcement learning
- DTransfer learning
Why this is the answer
Supervised learning requires labeled data; the model maps inputs to outputs using known target labels.
- Question 25Multiple Select
Which of the following are AWS managed AI services? (Select two.)
- AAmazon RekognitionCorrect answer
- BAmazon ComprehendCorrect answer
- CAmazon SageMaker
- DAWS Lambda
Why this is the answer
Rekognition and Comprehend are managed AI services; SageMaker is a platform for custom ML, and Lambda is serverless compute.
- Question 26Interactive
The following code snippet uses an AWS API to detect objects in an image. Which AWS service is being invoked?
- AAmazon RekognitionCorrect answer
- BAmazon Textract
- CAmazon Comprehend
- DAmazon Lex
Why this is the answer
The code calls the Rekognition API to detect labels (objects) in an image. Textract extracts text, Comprehend analyzes text, Lex builds chatbots.
- Question 27Single Choice
What is the first step in an ML pipeline according to best practices?
- AData collection
- BBusiness problem definitionCorrect answer
- CModel training
- DModel deployment
Why this is the answer
The first step is to define the business problem and success metrics before any data work begins.
- Question 28Multiple Select
Which two actions are examples of responsible AI practices? (Select two.)
- ABias detection with SageMaker ClarifyCorrect answer
- BManual hyperparameter tuning
- CModel explainability documentationCorrect answer
- DIncreasing model complexity to improve accuracy
Why this is the answer
Bias detection and explainability are core responsible AI practices; tuning and complexity are technical steps, not ethics-related.
- Question 29Interactive
The code trains a model using historical customer data with a column 'churned' as the target. What type of ML problem is this?
- ASupervised classificationCorrect answer
- BUnsupervised clustering
- CReinforcement learning
- DRegression
Why this is the answer
The target 'churned' is a binary label, so this is supervised classification. Regression predicts continuous values, not categories.
- Question 30Single Choice
Which AWS service is best suited for building a chatbot that understands user intents?
- AAmazon LexCorrect answer
- BAmazon Polly
- CAmazon Transcribe
- DAmazon Comprehend
Why this is the answer
Amazon Lex is for building conversational chatbots using ASR and NLU. Polly is text-to-speech, Transcribe is speech-to-text, Comprehend analyzes text.
- Question 31Multiple Select
Which two stages are part of the ML pipeline? (Select two.)
- AModel monitoringCorrect answer
- BData preparationCorrect answer
- CDatabase indexing
- DNetwork configuration
Why this is the answer
Data preparation and model monitoring are key ML pipeline stages; indexing and network config are infrastructure tasks not directly part of the pipeline.
ML Model Development
10 questions in this domain
- Question 32Single Choice
Which algorithm is best suited for predicting a continuous numerical target variable?
- ALinear RegressionCorrect answer
- BLogistic Regression
- CDecision Tree
- DK-Means
Why this is the answer
Linear Regression is used for continuous target variables; Logistic Regression is for classification.
- Question 33Multiple Select
Which of the following are supervised learning algorithms? (Select two.)
- ALinear RegressionCorrect answer
- BLogistic RegressionCorrect answer
- CK-Means Clustering
- DPrincipal Component Analysis
Why this is the answer
Linear Regression and Logistic Regression are supervised; K-Means and PCA are unsupervised.
- Question 34Interactive
Review the code snippet. Which parameter controls the type of instance used for training?
- Ainstance_typeCorrect answer
- Binstance_count
- Cvolume_size
- Drole
Why this is the answer
The instance_type parameter in SageMaker estimators determines the EC2 instance type for training.
- Question 35Single Choice
Which hyperparameter tuning strategy is most efficient for a large search space?
- AGrid Search
- BRandom Search
- CBayesian OptimizationCorrect answer
- DManual Search
Why this is the answer
Bayesian Optimization builds a surrogate model to choose promising hyperparameters, making it efficient for large spaces.
- Question 36Multiple Select
Which metrics are appropriate for evaluating a model on an imbalanced dataset? (Select two.)
- AAccuracy
- BPrecisionCorrect answer
- CRecallCorrect answer
- DF1 ScoreCorrect answer
Why this is the answer
Precision, Recall, and F1 are all suitable; but for two selections, Precision and Recall (or F1) are correct. The question expects any two of the three.
- Question 37Interactive
Examine the training output. What does the divergence of validation loss from training loss indicate?
- AOverfittingCorrect answer
- BUnderfitting
- CConvergence
- DData drift
Why this is the answer
When validation loss increases after a point while training loss decreases, the model is overfitting.
- Question 38Single Choice
Which SageMaker service should you use to orchestrate a multi-step ML workflow?
- ASageMaker Training Job
- BSageMaker PipelinesCorrect answer
- CSageMaker Experiments
- DAWS Step Functions
Why this is the answer
SageMaker Pipelines is designed for orchestrating end-to-end ML workflows as a DAG.
- Question 39Multiple Select
Which of the following are regularization techniques to prevent overfitting? (Select two.)
- AL1 RegularizationCorrect answer
- BDropoutCorrect answer
- CEarly Stopping
- DFeature Scaling
Why this is the answer
L1 Regularization and Dropout are regularization techniques; Early Stopping is a separate method, and Feature Scaling does not prevent overfitting.
- Question 40Interactive
The code shows a SageMaker Pipeline step. What does the 'Processing' step typically perform?
- AData transformation and feature engineeringCorrect answer
- BModel training
- CModel deployment
- DHyperparameter tuning
Why this is the answer
SageMaker Processing steps are used for data preprocessing, feature engineering, and evaluation.
- Question 41Single Choice
Which metric should you prioritize when false negatives are extremely costly?
- APrecision
- BRecallCorrect answer
- CAccuracy
- DSpecificity
Why this is the answer
Recall measures the proportion of actual positives correctly identified, minimizing false negatives.
Frequently asked questions
Everything about the AWS Certified AI Practitioner question bank and how to use it.