So you're aiming for the AWS Certified AI Practitioner (AIF-C01). It's a solid entry-level certification that validates your understanding of AI/ML concepts, AWS AI services, and responsible AI. But passing it isn't just about memorizing service names. You need to think through scenarios, understand trade-offs, and know when to use which tool.
Below are 10 representative questions that mirror the style and difficulty of the actual exam. Each one includes a full worked solution — not just the answer, but the reasoning you'd use to arrive at it. Use these to gauge your readiness and identify weak spots before you book the real thing.
1. Foundational Concepts: Supervised vs. Unsupervised Learning
Question: A data scientist wants to build a model that predicts customer churn based on historical data where each customer is labeled as "churned" or "retained." Which type of machine learning approach should they use?
- A) Unsupervised learning
- B) Reinforcement learning
- C) Supervised learning
- D) Semi-supervised learning
Solution: The correct answer is C) Supervised learning.
Why: The dataset includes labels (churned/retained), so the model learns from input-output pairs. Supervised learning is used for classification and regression tasks where ground truth labels are available. Unsupervised learning (A) would be for finding patterns without labels, like clustering. Reinforcement learning (B) involves an agent interacting with an environment to maximize rewards, not applicable here. Semi-supervised learning (D) uses a mix of labeled and unlabeled data, but the question states all customers are labeled.
2. AWS AI Services: Choosing the Right Tool
Question: A company needs to extract key information (like names, dates, and amounts) from thousands of scanned invoices and store it in a structured format. Which AWS service is best suited for this task?
- A) Amazon Textract
- B) Amazon Rekognition
- C) Amazon Comprehend
- D) Amazon Lex
Solution: The correct answer is A) Amazon Textract.
Why: Amazon Textract is specifically designed to extract text and structured data from documents, including forms and tables. It goes beyond simple OCR by understanding relationships and key-value pairs. Amazon Rekognition (B) is for image and video analysis, not documents. Amazon Comprehend (C) is for natural language processing (sentiment, entities) on text, not document parsing. Amazon Lex (D) is for building conversational interfaces.
3. Model Training: Overfitting and Mitigation
Question: A machine learning model performs extremely well on training data but poorly on unseen test data. What is this phenomenon called, and which technique can help mitigate it?
- A) Underfitting; increase model complexity
- B) Overfitting; apply regularization
- C) Bias; add more features
- D) Variance; use a simpler model
Solution: The correct answer is B) Overfitting; apply regularization.
Why: Overfitting occurs when the model memorizes the training data and fails to generalize. Regularization techniques (like L1/L2) penalize large weights, reducing complexity. Underfitting (A) is the opposite — poor performance on both training and test data. Bias (C) refers to systematic errors, not this scenario. Variance (D) is related but the best direct answer is overfitting with regularization as a standard mitigation.
4. Amazon SageMaker: Key Components
Question: Which Amazon SageMaker feature allows you to automate the process of finding the best hyperparameters for your model?
- A) SageMaker Ground Truth
- B) SageMaker Automatic Model Tuning
- C) SageMaker Experiments
- D) SageMaker Model Monitor
Solution: The correct answer is B) SageMaker Automatic Model Tuning.
Why: Automatic Model Tuning (also known as hyperparameter tuning) runs multiple training jobs with different hyperparameter combinations to find the best performing model. Ground Truth (A) is for labeling data. Experiments (C) is for tracking and comparing training runs. Model Monitor (D) is for detecting drift in deployed models.
5. Responsible AI: Bias and Fairness
Question: An AI model used for loan approvals is found to give different outcomes for applicants from different demographic groups. Which AWS service can help detect this bias?
- A) Amazon SageMaker Clarify
- B) AWS Audit Manager
- C) Amazon Macie
- D) AWS Identity and Access Management (IAM)
Solution: The correct answer is A) Amazon SageMaker Clarify.
Why: SageMaker Clarify provides tools to detect bias in training data and models, and also explains model predictions. Audit Manager (B) is for auditing AWS usage, not ML bias. Macie (C) is for data security and privacy (PII detection). IAM (D) is for access control.
6. Generative AI: Foundation Models
Question: A startup wants to build a chatbot that can answer customer queries in multiple languages. They plan to use a pre-trained foundation model and customize it with their own data. Which AWS service is designed for building and scaling generative AI applications with foundation models?
- A) Amazon Bedrock
- B) Amazon SageMaker JumpStart
- C) AWS Lambda
- D) Amazon Polly
Solution: The correct answer is A) Amazon Bedrock.
Why: Amazon Bedrock is a fully managed service that provides access to foundation models from various providers (AI21, Anthropic, Cohere, Meta, etc.) via API, and allows fine-tuning with your own data. SageMaker JumpStart (B) also offers pre-trained models but is more focused on the SageMaker ecosystem. Lambda (C) is for serverless compute, not directly for foundation models. Polly (D) is text-to-speech.
7. Model Deployment: Real-time vs. Batch
Question: A company needs to generate predictions for millions of customer records every night, but the results are not needed immediately. Which SageMaker inference option is most cost-effective?
- A) Real-time inference with a persistent endpoint
- B) Serverless inference
- C) Batch transform
- D) Asynchronous inference
Solution: The correct answer is C) Batch transform.
Why: Batch transform processes large datasets in chunks and is designed for offline predictions. It's more cost-effective than real-time endpoints (A) because you don't pay for idle capacity. Serverless (B) is for sporadic, low-latency requests. Asynchronous inference (D) is for near-real-time with queuing, but batch is best for nightly bulk processing.
8. AWS AI Services: Language Understanding
Question: Which AWS service can analyze customer feedback to determine whether the sentiment is positive, negative, or neutral?
- A) Amazon Transcribe
- B) Amazon Translate
- C) Amazon Comprehend
- D) Amazon SageMaker
Solution: The correct answer is C) Amazon Comprehend.
Why: Amazon Comprehend is a natural language processing (NLP) service that performs sentiment analysis, entity recognition, and key phrase extraction. Transcribe (A) converts speech to text. Translate (B) translates between languages. SageMaker (D) is a broader ML platform, but for a managed NLP service, Comprehend is the direct answer.
9. ML Pipeline: Data Preparation
Question: Before training a model, a data engineer needs to clean and transform raw data into a format suitable for training. Which SageMaker feature is specifically designed for building, running, and managing data processing jobs?
- A) SageMaker Processing
- B) SageMaker Feature Store
- C) SageMaker Data Wrangler
- D) SageMaker Pipelines
Solution: The correct answer is A) SageMaker Processing.
Why: SageMaker Processing allows you to run data processing jobs (like scikit-learn scripts) on managed infrastructure. Data Wrangler (C) is a visual interface for data preparation, but Processing is the underlying job engine. Feature Store (B) is for storing and sharing features. Pipelines (D) is for orchestrating the entire ML workflow, but Processing is the specific component for data processing.
10. Cost Optimization: ML Workloads
Question: A team is running multiple training jobs on SageMaker. They want to reduce costs without sacrificing performance. Which practice is most effective?
- A) Always use the largest instance type for faster training
- B) Use managed spot training for fault-tolerant workloads
- C) Keep endpoints running 24/7 for immediate access
- D) Store all training data in Amazon S3 Standard storage
Solution: The correct answer is B) Use managed spot training for fault-tolerant workloads.
Why: Spot instances offer up to 90% discount and are ideal for training jobs that can be interrupted and resumed. Using the largest instance (A) increases cost. Keeping endpoints running (C) incurs idle charges. S3 Standard (D) is not a cost-saving measure for training data; you might use S3 Intelligent-Tiering or Glacier for infrequent access.
Take a free AWS Certified AI Practitioner demo mock to find out where you stand: Try the demo →
How to Use These Sample Questions
Don't just read the explanations — actually attempt each question on your own first. Write down your answer, then compare with the solution. If you get it wrong, note the reason. Are you confusing service names? Do you need to revisit the basics of ML? This diagnostic approach will help you focus your study time.
What to Study Next
- AWS AI Services Deep Dive: Understand the core use cases for Textract, Rekognition, Comprehend, Polly, Transcribe, Translate, and Lex. Know the difference between AI services (managed) and ML services (SageMaker).
- SageMaker Fundamentals: Familiarize yourself with the key components: Ground Truth, Processing, Training, Tuning, Deployment, and Monitoring.
- Responsible AI: Study bias, fairness, explainability, and governance. AWS has specific tools like Clarify and Model Monitor.
- Generative AI: Get comfortable with Amazon Bedrock, foundation models, and prompt engineering basics.
Where to Next?
- Take the full AWS Certified AI Practitioner mock test to simulate the real exam experience.
- Explore our AWS AI Practitioner study guides for in-depth coverage of every domain.
- See AWS Certified AI Practitioner mock-test packs and pricing to choose a plan that fits your budget.
See AWS Certified AI Practitioner mock-test packs and pricing: View plans →