AIF-C01 Practice Exam 55 — 20 Questions

Question 1

A financial services company is preparing to train a machine learning model on customer transaction data. The data science team must address data quality concerns before training, because poor data directly harms model performance. Which TWO practices best improve the quality of the training data? (Choose two.)

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Handle missing values by imputing them or removing affected records after analysis. Improving data quality before training centers on correcting defects in the data itself. Handling missing values prevents bias and instability, while scaling numeric features ensures algorithms treat inputs comparably. Hyperparameter changes, row duplication, and deleting predictive features do not repair data defects and in several cases worsen results.

Answer

Add more duplicate rows to increase the effective dataset size

Answer

Increase the learning rate so the model converges faster on noisy data

Answer

Remove all features that have any correlation with the target variable

Question 2

A startup is preparing a dataset to train a supervised machine learning model and wants to follow sound data preparation practices. Which TWO activities are appropriate parts of preparing training data? (Choose two.)

Accepted Answer

Splitting the dataset into training, validation, and test subsets before tuning the model.. Sound preparation centers on protecting evaluation integrity and improving data quality: separating train, validation, and test subsets keeps performance estimates honest, while removing duplicates and fixing inconsistent labels prevents leakage and contradictory signals. Tuning on the test set, deleting all incomplete rows, and duplicating examples either leak information, discard useful data, or add no new information.

Answer

Adding synthetic copies of the same training examples until the dataset reaches a target size.

Answer

Tuning the model's hyperparameters repeatedly against the test set until the score improves.

Answer

Deleting all records that contain any missing values to guarantee a clean dataset.

Question 3

A solutions architect is explaining why a foundation model can answer questions about topics it was never explicitly programmed for. Which characteristic of generative AI best explains this behaviour?

Accepted Answer

The model learned statistical patterns and relationships from large-scale training data, allowing it to generalize to new prompts.. Generative models generalize because training over massive datasets encodes statistical patterns and relationships among tokens into the model weights. At inference the prompt is processed against those learned parameters, enabling coherent responses to inputs never seen in training. Retrieval, verbatim lookup, and per-prompt retraining all describe different mechanisms that do not explain inherent generalization.

Answer

The model retrieves live answers from a search index at inference time.

Answer

The model re-trains itself on each user prompt before producing a response.

Answer

The model stores every training example verbatim and looks up the closest match when prompted.

Question 4

A company is using Amazon Bedrock to power a chatbot that provides customer support. The security team wants to ensure that the chatbot does not generate responses that include profanity, hate speech, or prompts that attempt to bypass safety filters (jailbreak attempts). They also want to log any blocked interactions for review. Which AWS service or feature should be used to meet these requirements?

Accepted Answer

Amazon Bedrock Guardrails with content filters and prompt attack detection, and enable model invocation logging to capture blocked interactions.. The requirements are to block harmful content and jailbreak attempts in a Bedrock-powered chatbot, and to log blocked interactions. Amazon Bedrock Guardrails offers content filters for profanity, hate speech, and more, plus prompt attack detection. Enabling model invocation logging captures all interactions, including those blocked by Guardrails, for later review. This is a managed solution that directly addresses the needs without custom development.

Answer

Amazon Comprehend for sentiment analysis and Amazon Macie for detecting sensitive data, with alerts sent to Amazon SNS.

Answer

Amazon SageMaker Clarify for bias detection and Amazon Augmented AI (A2I) for human review, with results logged to Amazon S3.

Answer

AWS WAF with managed rules for bot control and a custom Lambda function to inspect responses for profanity, logging to Amazon CloudWatch.

Question 5

A government agency trains models on highly sensitive data inside Amazon SageMaker. The security policy states that training data and model artifacts must never be accessible over the public internet and that traffic to AWS services must stay within the agency's Amazon VPC. Which combination should the agency implement?

Accepted Answer

Run training jobs with network isolation enabled and access AWS services through VPC interface endpoints powered by AWS PrivateLink.. The policy has two distinct demands: no public internet access from the training environment and private connectivity to AWS services. Network isolation removes outbound network access entirely, and VPC interface endpoints route service calls over private addresses. IP conditions, NAT-routed public subnets, and IAM-only controls each address authorization or partial routing rather than the complete private-path requirement.

Answer

Place the training job in a public subnet with a NAT gateway and restrict security group egress to AWS service prefix lists.

Answer

Attach an S3 bucket policy that denies requests where aws:SourceIp is outside the agency CIDR range and enable default encryption.

Answer

Use SageMaker Studio in a private subnet and rely on AWS Identity and Access Management policies to deny access to non-approved services.

Question 6

A solutions architect is explaining the concept of a foundation model to a non-technical stakeholder. The stakeholder asks what distinguishes a foundation model from a traditional task-specific machine learning model. Which statement best describes a foundation model?

Accepted Answer

It is a large model pre-trained on broad data that can be adapted to many downstream tasks.. A foundation model is characterized by large-scale pre-training on broad data, which yields general capabilities that can be adapted to numerous downstream tasks via prompting, fine-tuning, or other techniques. This generality is what differentiates it from task-specific models built for a single narrow purpose, and it is the essence the architect needs to convey.

Answer

It is a model trained exclusively on a company's proprietary data for one specific use case.

Answer

It is a small model optimized to run only on edge devices with limited compute.

Answer

It is a rule-based system that follows hand-coded logic to produce deterministic outputs.

Question 7

A media company wants to build a system that generates short promotional descriptions for its articles. The team has no labeled dataset of article-summary pairs but has a large corpus of published articles and descriptions. They want to leverage a pretrained foundation model and adapt it to their domain with minimal labeling effort. Which approach best fits this scenario?

Accepted Answer

Fine-tune a pretrained foundation model on the company's article and description corpus for text generation.. The company needs generated text and already holds a domain corpus of articles paired with descriptions. Fine-tuning a pretrained foundation model adapts broad language ability to the company's style with modest additional data, which matches the stated preference for minimal labeling. Classification, clustering, and from-scratch reinforcement learning either produce the wrong output type or discard the pretrained model advantage.

Answer

Use reinforcement learning with reader click rewards to train a description generator from scratch.

Answer

Train a supervised classification model that assigns each article to a predefined topic category.

Answer

Apply unsupervised clustering to group articles by topic and use cluster IDs as descriptions.

Question 8

A data science intern is reading about the transformer architecture that underpins most modern large language models. The intern asks which mechanism allows a transformer to weigh the relevance of every other word in a sentence when encoding a given word, regardless of how far apart the words appear. Which mechanism should you name?

Accepted Answer

Self-attention. Self-attention is the operation that lets each token build its representation from all other tokens in the sequence with learned weights, which is what removes the long-distance constraint of sequential models. Positional encoding, tokenization, and recurrent state propagation all play supporting or unrelated roles, but none of them performs the relevance weighting described in the scenario.

Answer

Positional encoding

Answer

Tokenization

Answer

Recurrent state propagation

Question 9

A retail bank has millions of unlabeled customer transaction records and wants to discover natural groupings of spending behavior without defining any categories in advance. The data science team plans to use an unsupervised learning approach. Which technique is designed for this goal?

Accepted Answer

K-means clustering, which partitions records into groups so that members of a group are more similar to each other.. The bank has no predefined categories and wants to uncover structure in unlabeled data, which is the defining use case for unsupervised learning. K-means clustering groups similar records together so the team can inspect and name the resulting segments afterward, turning raw transaction behavior into actionable customer groupings.

Answer

Linear regression, which models the relationship between a continuous target and one or more input features.

Answer

Logistic regression, which estimates the probability that a record belongs to a particular class.

Answer

A decision tree classifier, which splits data using feature thresholds to predict a labeled category.

Question 10

A media company stores thousands of hours of unlabeled video footage and wants to build a searchable index that lets editors retrieve clips by describing their content in natural language. The team has no annotated dataset and no budget to label one. Which machine learning approach is the MOST appropriate starting point?

Accepted Answer

Use a pretrained multimodal embedding model to encode video frames and text into a shared vector space, then retrieve clips by similarity.. With no labels and a need to search footage using free-form language, a pretrained multimodal embedding model is the most practical foundation. It places video and text in a shared vector space so that descriptive queries match relevant clips by similarity, enabling semantic retrieval without any custom annotation effort.

Answer

Train a supervised video classifier from scratch using the footage, treating each video file as its own class.

Answer

Apply k-means clustering to raw pixel values of every frame to produce searchable cluster identifiers.

Answer

Build a reinforcement learning agent that learns to select the clip most likely to satisfy an editor through trial and error.

Question 11

A media company stores 40 TB of raw video footage in Amazon S3 and wants to automatically detect scene boundaries, identify on-screen text, and flag unsafe frames without building custom computer vision models. Which AWS service should they use?

Accepted Answer

Amazon Rekognition Video. Amazon Rekognition Video is purpose-built for analyzing video stored in S3 or streamed in real time, offering scene detection, text detection, content moderation, and activity recognition through managed APIs. Because the company wants those visual capabilities without training custom models, the managed video analysis service is the correct fit. The other services address speech, text, or audio generation rather than visual video understanding.

Answer

Amazon Comprehend

Answer

Amazon Polly

Answer

Amazon Transcribe

Question 12

A financial services firm is evaluating foundation models for a customer-facing assistant. Compliance requires that prompts and completions never leave the firm's own AWS account boundary. Which characteristic of a foundation model deployment should the team evaluate FIRST against this constraint?

Accepted Answer

Whether the model is consumed as a fully managed service or deployed into infrastructure the firm controls.. Data-residency and account-boundary requirements are determined by where inference actually executes, so the deployment and consumption model must be assessed before capability metrics. Fully managed endpoints process requests on provider-managed infrastructure, while models deployed into the firm's own account keep prompt and completion data inside that boundary.

Answer

The number of parameters in the model and its published benchmark scores.

Answer

The maximum number of tokens the model can accept in a single request.

Answer

The licensing terms that govern commercial redistribution of the model weights.

Question 13

A company is using Amazon Bedrock to build an AI assistant that accesses internal knowledge bases. The security team wants to ensure that the assistant only retrieves documents that the requesting user is authorized to view, based on their department. Which approach should be used to enforce this fine-grained access control?

Accepted Answer

Implement document-level access control using metadata filtering in Amazon Bedrock Knowledge Bases. Amazon Bedrock Knowledge Bases supports metadata filtering, which allows documents to be tagged with attributes such as department. When a user queries the knowledge base, the application can pass a filter based on the user's department, ensuring only authorized documents are retrieved. This enforces fine-grained access control at the document level.

Answer

Use Amazon Bedrock Guardrails to filter responses by department

Answer

Store all documents in a single S3 bucket with a bucket policy allowing only the assistant's IAM role

Answer

Configure IAM policies to restrict access to the Bedrock model

Question 14

A machine learning team notices their model performs excellently on the training dataset but poorly on new, unseen data. They want to reduce this gap without collecting more data. Which action most directly addresses the problem?

Accepted Answer

Apply regularization techniques such as L2 penalty or dropout. The described pattern is classic overfitting: strong training performance with weak performance on unseen data. Regularization such as L2 penalties or dropout constrains the model and improves generalization. Adding capacity, training longer, or discarding validation all worsen or conceal the issue instead of correcting it.

Answer

Increase model complexity by adding more layers and parameters

Answer

Train for many more epochs until training loss approaches zero

Answer

Remove the validation dataset and evaluate only on training data

Question 15

A media company is evaluating foundation models for a generative AI application that produces image captions and short video summaries. The team must balance output quality, latency, and operational cost. Which TWO considerations are most important when selecting a foundation model for this multimodal task? (Choose two.)

Accepted Answer

Whether the model supports the required input modalities, such as images and video frames.. For a multimodal captioning and summarization workload, the model must accept images and video frames as input, so modality support is a prerequisite. Because the team must also balance latency and cost, evaluating inference speed and pricing against throughput needs is equally critical. Together these ensure the model can handle the data and remain practical to operate at scale.

Answer

Whether the model can generate outputs in multiple human languages simultaneously.

Answer

Whether the model was trained using a specific programming language's libraries.

Answer

Whether the model's license permits the company's intended commercial distribution of outputs.

Question 16

A media company wants a generative AI assistant that drafts scripts in a distinctive house style. The team has several thousand pages of approved scripts but no labeled input-output pairs, and they want to adapt an existing foundation model rather than train one from scratch. Which approach best matches their data and goal?

Accepted Answer

Continued pre-training on the unlabeled script corpus to adapt the model to the house style. Continued pre-training is designed for exactly this situation: a large body of unlabeled domain text and a desire to shift an existing foundation model toward that domain. Supervised fine-tuning and RLHF both need labels the team does not have, and few-shot prompting cannot absorb thousands of pages of stylistic signal into the model itself.

Answer

Reinforcement learning from human feedback using preference rankings of generated scripts

Answer

Supervised fine-tuning on prompt-completion pairs extracted from the scripts

Answer

Few-shot prompting with three example scripts inserted into each request

Question 17

A media company wants to create an internal tool that drafts short news summaries from press releases. They have no machine learning engineers, no labeled training data, and need a working prototype within two weeks. Which approach best describes how they should build this capability using generative AI concepts on AWS?

Accepted Answer

Use a pretrained foundation model through an API and steer its behavior with prompt engineering and few-shot examples.. Foundation models are pretrained on broad corpora and can be adapted to a new task through prompting rather than retraining. Supplying instructions plus a few example summaries in context lets a team with no ML specialists and no labeled dataset produce a functional summarizer within days. Training from scratch or using a vision service does not fit the constraint or the modality.

Answer

Build a rules-based extractive summarizer that selects the first and last sentence of each press release.

Answer

Deploy an Amazon Rekognition custom labels project and post-process its output into summary sentences.

Answer

Train a small transformer model from scratch on the company's archive of press releases using Amazon SageMaker training jobs.

Question 18

A media company is experimenting with an Amazon Bedrock text model to draft short news summaries. The team notices that when they ask the same question twice, the model returns noticeably different wording each time, and sometimes the summary drifts off topic. They want more deterministic, focused responses without retraining the model. Which combination of inference parameters should they adjust to reduce randomness and keep the output on topic?

Accepted Answer

Decrease the temperature and set a lower topP value.. Randomness and topical drift in foundation model output are governed by sampling parameters. Temperature scales the probability distribution, and topP restricts sampling to the smallest set of tokens whose cumulative probability exceeds the threshold. Lowering both concentrates generation on the most likely tokens, yielding more consistent and focused summaries without any model retraining.

Answer

Increase the temperature and increase the topP value.

Answer

Enable streaming responses and lower the topK value to zero.

Answer

Increase maxTokens and decrease the stop sequence length.

Question 19

A retail bank is deploying an Amazon SageMaker model that recommends credit limit increases to existing cardholders. The bank's responsible AI review board requires that the model's decisions be explainable to customers who request an adverse action notice, and that the team be able to detect whether any single input feature is disproportionately driving predictions. Which TWO capabilities should the team implement to meet these requirements? (Choose two.)

Accepted Answer

Use SageMaker Clarify explainability to generate SHAP-based feature attributions for individual predictions. SageMaker Clarify provides both SHAP-based per-prediction explanations and bias metrics such as disparate impact and feature importance. Together they let the bank produce customer-facing adverse action reasons and detect whether a particular feature disproportionately drives decisions. Model Monitor, Debugger, and load balancer controls address drift, training internals, and traffic security rather than explainability or fairness.

Answer

Enable SageMaker Debugger to capture tensor values during training

Answer

Deploy the model behind an Application Load Balancer with AWS WAF for request filtering

Answer

Configure SageMaker Model Monitor to track feature drift against a baseline distribution

Question 20

A financial services firm wants an internal assistant that answers employee questions about company travel policy in natural language. The policy documents change frequently, and the firm requires answers to cite the specific policy section used. Which approach BEST meets these requirements?

Accepted Answer

Use retrieval-augmented generation, where relevant policy passages are retrieved and supplied to the model to ground its answer and source citation.. Retrieval-augmented generation grounds each answer in freshly retrieved policy passages, so responses reflect the current documents and can point to the exact section used. Because updates only require re-indexing changed content rather than retraining a model, it also handles the frequently changing policy corpus efficiently.

Answer

Deploy a keyword search engine that returns matching policy paragraphs and rely on employees to interpret the results themselves.

Answer

Fine-tune a large language model on the full policy corpus so it memorizes the documents and answers from memory.

Answer

Train a small classification model that maps each employee question to one of the fixed policy section titles.