Skip to main content
Etalorix

AWS Certified AI Practitioner (AIF-C01): sample questions with answers

20 sample questions, each with the correct answer and an explanation. They come from the question bank and follow the exam format.

Questions are written from samples prepared by specialists in the field and follow the official exam format.

  1. 1. Why are documents split into chunks before embedding in a RAG pipeline?

    1. A) To reduce model size
    2. B) So relevant passages fit in the context window and retrieval is more precise
    3. C) To make them encrypted
    4. D) To delete them
    Show the answer and explanation

    Correct answer: B

    So passages fit the context and retrieval is precise.

  2. 2. Semantic search differs from keyword search because it:

    1. A) Is always slower
    2. B) Cannot use vectors
    3. C) Only matches exact words
    4. D) Matches meaning using embeddings, not just exact words
    Show the answer and explanation

    Correct answer: D

    In short: It matches meaning using embeddings.

  3. 3. Extracting data from invoices and then summarising them would best combine:

    1. A) Amazon Forecast and Amazon Personalize
    2. B) Amazon Textract and a foundation model in Amazon Bedrock
    3. C) Amazon Rekognition and Amazon SQS
    4. D) Amazon Polly and Amazon Lex
    Show the answer and explanation

    Correct answer: B

    In short: Amazon Textract and a Bedrock model.

  4. 4. Which metric is commonly used to evaluate text summarisation?

    1. A) ROUGE
    2. B) Mean absolute error
    3. C) AUC
    4. D) RMSE
    Show the answer and explanation

    Correct answer: A

    In short: ROUGE. This is the correct answer.

  5. 5. When is human evaluation of model outputs most valuable?

    1. A) Only after deployment ends
    2. B) When judging qualities such as helpfulness, tone and accuracy that automatic metrics miss
    3. C) Never, metrics are enough
    4. D) Only for image models
    Show the answer and explanation

    Correct answer: B

    In short: To judge helpfulness, tone and accuracy.

  6. 6. Asking the model to 'think step by step' is an example of:

    1. A) Fine-tuning
    2. B) Chain-of-thought prompting
    3. C) Tokenisation
    4. D) Prompt injection
    Show the answer and explanation

    Correct answer: B

    In short: Chain-of-thought prompting. This is the correct answer.

  7. 7. Including several input–output examples in the prompt is:

    1. A) Distillation
    2. B) Zero-shot prompting
    3. C) Continued pre-training
    4. D) Few-shot prompting
    Show the answer and explanation

    Correct answer: D

    In short: Few-shot prompting. This is the correct answer.

  8. 8. Which AWS service can serve as a vector store for Knowledge Bases for Amazon Bedrock?

    1. A) AWS Lambda
    2. B) Amazon OpenSearch Serverless
    3. C) Amazon CloudFront
    4. D) Amazon SQS
    Show the answer and explanation

    Correct answer: B

    In short: Amazon OpenSearch Serverless. This is the correct answer.

  9. 9. Which Bedrock feature provides a managed RAG workflow over a company's documents?

    1. A) AWS Glue
    2. B) Bedrock Guardrails
    3. C) Knowledge Bases for Amazon Bedrock
    4. D) Amazon Polly
    Show the answer and explanation

    Correct answer: C

    In short: Knowledge Bases for Amazon Bedrock.

  10. 10. Predicting whether an email is spam or not spam is a:

    1. A) Time-series forecast
    2. B) Clustering problem
    3. C) Regression problem
    4. D) Binary classification problem
    Show the answer and explanation

    Correct answer: D

    In short: A binary classification problem.

  11. 11. Which tool lets business analysts build ML models without writing code?

    1. A) Amazon SageMaker Canvas
    2. B) Amazon EC2
    3. C) Amazon SageMaker Debugger
    4. D) AWS Lambda
    Show the answer and explanation

    Correct answer: A

    In short: Amazon SageMaker Canvas. This is the correct answer.

  12. 12. A fraud model flags 100 transactions, of which 80 are actually fraudulent. What is its precision?

    1. A) 80%
    2. B) Cannot be computed
    3. C) 100%
    4. D) 20%
    Show the answer and explanation

    Correct answer: A

    Calculation: 80 / 100 = 80%. This is the correct answer.

  13. 13. Grouping customers into segments without predefined labels is an example of:

    1. A) Supervised classification
    2. B) Regression
    3. C) Unsupervised learning (clustering)
    4. D) Reinforcement learning
    Show the answer and explanation

    Correct answer: C

    Clustering without labels is unsupervised learning.

  14. 14. Predicting the price of a house from its features is a:

    1. A) Clustering problem
    2. B) Anomaly detection only
    3. C) Regression problem
    4. D) Classification problem
    Show the answer and explanation

    Correct answer: C

    In short: A regression problem. This is the correct answer.

  15. 15. Which step usually comes first when building a generative AI application?

    1. A) Deploy to production
    2. B) Buy GPUs
    3. C) Define the use case and success criteria
    4. D) Fine-tune a model
    Show the answer and explanation

    Correct answer: C

    In short: Define the use case and success criteria.

  16. 16. An embedding is:

    1. A) A password
    2. B) A compressed image file
    3. C) A numerical vector that represents the meaning of text or other data
    4. D) A type of GPU
    Show the answer and explanation

    Correct answer: C

    In short: A numerical vector representing meaning.

  17. 17. Which is a responsible AI consideration about sustainability?

    1. A) Ignoring energy use
    2. B) Always using the largest model
    3. C) Choosing efficient model sizes and infrastructure to reduce energy use
    4. D) Training repeatedly without need
    Show the answer and explanation

    Correct answer: C

    Choose efficient models and infrastructure.

  18. 18. Which practice is part of good generative AI governance?

    1. A) Letting every team choose any tool without rules
    2. B) Defining approved use cases, data policies, roles and review processes
    3. C) Disabling audits
    4. D) Skipping risk assessment
    Show the answer and explanation

    Correct answer: B

    Define approved uses, data policies, roles and reviews.

  19. 19. Why is data lineage important for AI governance?

    1. A) It reduces token costs
    2. B) It shows where training data came from and how it was transformed, supporting audits
    3. C) It speeds up inference
    4. D) It replaces encryption
    Show the answer and explanation

    Correct answer: B

    It shows data origin and transformation for audits.

  20. 20. Data poisoning is an attack in which:

    1. A) Malicious data is inserted into training data to corrupt model behaviour
    2. B) A model is copied
    3. C) A user forgets a password
    4. D) Prompts are encrypted
    Show the answer and explanation

    Correct answer: A

    In short: Malicious data corrupts training.