AI Fundamentals and Literacy

Question 1

A logistics company stores thousands of standardized scanned shipment forms in Amazon S3. Each form contains a tracking number, shipment date, and delivery address, but minor layout differences exist between suppliers. The company wants an automated solution without collecting a training dataset or operating model infrastructure. Which solution best satisfies the requirements?

A. Send each document to a general-purpose foundation model and ask it to infer the tracking number, date, and address.

B. Use Amazon Textract to extract text and form fields from documents stored in Amazon S3, invoking it through AWS Lambda and routing low-confidence results to Amazon A2I.

C. Build and train a custom document model with Amazon SageMaker, including a labeling project and dedicated inference endpoint.

D. Use AWS Lambda regular expressions to locate fields based on fixed character positions in each scanned image.

Question 2

A media company classifies customer feedback into billing, playback, account, and content categories. The language and category definitions change frequently. The team has some labeled historical feedback in Amazon S3 and can use Amazon Comprehend, Amazon SageMaker, and Amazon CloudWatch. Which factors should influence the decision between a managed Amazon Comprehend capability and a custom Amazon SageMaker model? Select all that apply.

A. Compare the required customization and update frequency with the operational cost of training, deploying, and monitoring a custom model.

B. Select a custom model whenever the team wants maximum control, regardless of managed-service coverage or maintenance cost.

C. Determine whether Amazon Comprehend satisfies at least 80 percent of the accuracy, language, throughput, and category requirements.

D. Assume a custom model is always superior because it can be retrained whenever categories change.

E. Assess the volume, representativeness, labeling consistency, and freshness of historical training data.

Question 3

A manufacturer stores temperature readings, maintenance dates, and machine-failure labels in Amazon S3. A data team uses AWS Glue Data Catalog and Amazon SageMaker to build a model that predicts whether a machine will fail within seven days. Which statement correctly describes the workflow?

A. Current sensor readings are used to train the model, and historical labeled records are used only after deployment for inference.

B. The trained model is a static rule set that produces predictions without requiring new input data.

C. Historical records containing failure labels are used during training, and current sensor readings are supplied during inference to produce a seven-day failure prediction.

D. Predictions are generated during training without input features because the failure label alone determines the result.

Question 4

A grocery chain stores transaction totals, product IDs, and quantities in relational tables in Amazon RDS. It also stores receipt images and customer-service call recordings in Amazon S3. The company uses AWS Glue and Amazon Transcribe before downstream analytics. Which statements correctly classify and handle this data? Select all that apply.

A. Images and audio commonly require preprocessing or specialized services such as Amazon Transcribe before modeling. ...