AWS Certified AI Practitioner (AIF-C01) practice questions

Question 1 · Fundamentals of AI and ML

Nordhaven Freight runs a fleet of refrigerated trucks. Engineers log 14 sensor readings per truck per minute and have three years of historical records showing which compressor units eventually failed and on what date. Leadership wants a system that flags a compressor as likely to fail within the next 30 days. Which machine learning approach fits this problem?

1. A.Unsupervised learning, because the sensor readings arrive as a continuous unlabeled stream
2. B.Supervised learning, because the historical records provide labeled failure outcomes to train against
3. C.Semi-supervised learning, because sensor data is numeric rather than categorical
4. D.Reinforcement learning, because the trucks operate in a changing physical environment

Question 2 · Fundamentals of GenAI

A solutions architect at Fennimore Media is explaining why the company's new drafting assistant is built on a foundation model rather than a task-specific model. Which characteristic of foundation models best supports that decision?

1. A.They run only on customer-owned hardware and cannot be accessed through an API
2. B.They require the customer to supply the full training corpus before first use
3. C.They are trained on a narrow labeled dataset for one predefined output class
4. D.They are pretrained on broad data at scale and can be adapted to many downstream tasks

Question 3 · Applications of Foundation Models

Ashgrove Legal wants a foundation model to consistently reply in the firm's formal house style and to use its internal terminology, and it has 2,000 curated prompt-and-response pairs written by senior associates. Which customization approach fits this goal?

1. A.Enabling provisioned throughput on the base model endpoint
2. B.Increasing the top-k sampling parameter on each request
3. C.Adding more documents to a vector store for retrieval
4. D.Fine-tuning the model on the curated prompt-response pairs

Question 4 · Guidelines for Responsible AI

During testing of a resume screening tool, an audit shows the model advances candidates from one graduate school at a much higher rate than equally qualified candidates from other schools, tracing back to the historical hiring data used for training. Which responsible AI concern does this represent?

1. A.Excessive inference latency
2. B.Weak encryption of the model artifacts
3. C.Insufficient model context window
4. D.Bias inherited from the training data

Question 5 · Security, Compliance, and Governance for AI Solutions

Under the AWS shared responsibility model, a customer builds an application on Amazon Bedrock. Which task belongs to the customer rather than to AWS?

1. A.Applying firmware updates to the underlying accelerator hardware
2. B.Maintaining physical security of the data centers running the service
3. C.Configuring IAM policies that determine which principals may invoke which models
4. D.Patching the hypervisor hosting the managed inference fleet

Question 6 · Fundamentals of AI and ML

A data scientist at Kestrel Analytics trains a customer-churn model that scores 98.4% accuracy on the training set but only 71% on a held-out test set. Which description best characterizes what has happened and the most direct remedy?

1. A.The model is overfitting; the team should apply regularization or gather more diverse training data
2. B.The training data is imbalanced; the team should switch the loss function to mean squared error
3. C.The model is underfitting; the team should add more layers and train for more epochs
4. D.The features are poorly scaled; the team should convert all numeric columns to text embeddings

Question 7 · Fundamentals of GenAI

An engineer at Sandrick Logistics observes that a generative model confidently states that the company's fictional "Route Optimizer 7" product supports rail freight, which it does not. What is this behavior called, and which mitigation most directly reduces it?

1. A.Latency spike; move the endpoint to a larger instance type
2. B.Overfitting; shrink the model's context window
3. C.Hallucination; ground responses in retrieved authoritative documents
4. D.Tokenization error; switch the request to a different character encoding

Question 8 · Applications of Foundation Models

A developer at Coastwise Travel needs the assistant to look up live flight status by calling an internal REST API and then answer the customer in natural language. Which Amazon Bedrock capability is designed for orchestrating that kind of multi-step task with external API calls?

1. A.Amazon Bedrock model evaluation jobs
2. B.Amazon Bedrock Agents
3. C.Amazon Bedrock Guardrails
4. D.Amazon Bedrock provisioned throughput