Key details for this exam, checked against the published exam outline
Each question shows the correct answer and an explanation of why it is right
A company has a Retrieval Augmented Generation (RAG) application that uses a vector database to store embeddings of documents. The company must migrate the application to AWS and must implement a solution that provides semantic search of text files. The company has already migrated the text repository to an Amazon S3 bucket.
Which solution will meet these requirements?
The key requirement is semantic search over text documents that already reside in Amazon S3. AWS provides Amazon Kendra, a fully managed service specifically designed for semantic and natural language search across unstructured text.
Amazon Kendra natively supports S3 connectors, which can ingest documents directly from an S3 bucket, automatically process the text, generate embeddings, and index the content for semantic retrieval. This removes the need for the company to manage embedding generation, vector storage, or similarity search infrastructure. Queries can be expressed in natural language, making Kendra well suited for RAG-style applications.
Option A and B require building and maintaining a custom embedding pipeline and do not provide a true vector similarity search engine using SQL. SageMaker Feature Store is not intended to function as a vector database for semantic search.
Option D is incorrect because Amazon Textract is an OCR service for extracting text from scanned documents and images; it does not support semantic search.
Therefore, ingesting documents using the Amazon Kendra S3 connector and querying Kendra is the correct and AWS-recommended solution.
An ML engineer is tuning an image classification model that shows poor performance on one of two available classes during prediction. Analysis reveals that the images whose class the model performed poorly on represent an extremely small fraction of the whole training dataset.
The ML engineer must improve the model's performance.
Which solution will meet this requirement?
This problem describes severe class imbalance in an image classification task, where the minority class has poor predictive performance. In such cases, accuracy is a misleading metric, because a model can achieve high accuracy by predicting only the majority class. AWS ML best practices recommend using F1 score, which balances precision and recall and is more appropriate for imbalanced classification problems.
To improve performance on the minority image class, image augmentation is the preferred approach. Augmentation techniques---such as rotation, cropping, flipping, and brightness adjustment---create realistic new training examples while preserving semantic meaning. AWS documentation recommends augmentation for computer vision workloads to improve generalization without collecting new data.
SMOTE (Options C and D) is designed for tabular data, not image data, and generating synthetic pixel-level images using SMOTE is not appropriate or supported in typical computer vision pipelines.
Option A is incorrect because optimizing for accuracy does not address minority-class performance. Option D is incorrect because SMOTE is unsuitable for images.
Therefore, optimizing for F1 score and using image augmentation on the minority class is the correct solution.
A company wants to use large language models (LLMs) supported by Amazon Bedrock to develop a chat interface for internal technical documentation.
The documentation consists of dozens of text files totaling several megabytes and is updated frequently.
Which solution will meet these requirements MOST cost-effectively?
AWS recommends Retrieval Augmented Generation (RAG) using Amazon Bedrock knowledge bases as the most cost-effective solution for incorporating frequently updated documents into LLM-powered applications.
A Bedrock knowledge base allows the company to store documents in Amazon S3, index them using vector embeddings, and retrieve relevant context dynamically at inference time. This approach eliminates the need for retraining or fine-tuning the model when documents change.
Training or fine-tuning an LLM is expensive, time-consuming, and unnecessary for frequently changing data. Bedrock guardrails are designed for safety and policy enforcement, not knowledge integration.
AWS documentation explicitly states that knowledge bases are the preferred method for dynamic, updatable enterprise content in chat applications.
Therefore, Option D is the correct and most cost-effective solution.
A company is developing ML models by using PyTorch and TensorFlow estimators with Amazon SageMaker AI. An ML engineer configures the SageMaker AI estimator and now needs to initiate a training job that uses a training dataset.
Which SageMaker AI SDK method can initiate the training job?
In the Amazon SageMaker Python SDK, the fit() method is used to start a training job after an estimator has been configured. AWS documentation explicitly states that once an estimator (such as PyTorch or TensorFlow) is defined with parameters like instance type, framework version, and hyperparameters, the fit() method is responsible for launching the training process.
The fit() method accepts the training data location (commonly an Amazon S3 URI) and initiates the managed training job on SageMaker infrastructure. SageMaker then provisions the required compute resources, stages the data, executes the training script, and stores model artifacts in Amazon S3.
The create_model() method is used after training to create a SageMaker model object from trained artifacts. The deploy() method deploys a trained model to an endpoint for inference. The predict() method is used only after deployment to request predictions from an endpoint.
AWS documentation clearly separates these lifecycle steps and identifies fit() as the correct method to initiate training.
Therefore, Option A is the correct and AWS-verified answer.
A company is using an ML model to classify motion in videos. The data is stored in MP4 format in Amazon S3. When the company created the model, the company needed 4 months to label all the video frames.
The company needs to retrain the model with an existing training workflow in Amazon SageMaker AI. An ML engineer must implement a solution that decreases the labeling time.
Which solution will meet these requirements?
Option A is correct because Amazon SageMaker Ground Truth is the AWS service specifically designed to build labeled datasets for machine learning, including video classification and video frame labeling use cases. AWS documentation states that Ground Truth can take video files in MP4 format stored in Amazon S3, automatically connect to that data, and extract video frames for labeling. This directly matches the scenario in the question, where the source data is MP4 in S3 and the company needs to relabel or expand training data for retraining an existing SageMaker workflow.
Ground Truth is also the best choice because AWS documents explain that it can reduce labeling time and cost through automated data labeling (active learning) for supported built-in task types. Instead of relying only on manual human labeling for every item, Ground Truth can use ML-assisted labeling to lower the total amount of human work needed, which is exactly what the company wants after previously spending four months labeling frames.
The other options do not fit as well. JumpStart provides pretrained models and solution templates, but AWS documentation does not position it as the service for running a video-frame labeling operation. Data Wrangler is for data preparation, transformation, and feature engineering, not human labeling of video frames. Amazon A2I with Rekognition is mainly documented for human review workflows such as Rekognition moderation outputs, not as the primary service for large-scale MP4 video-frame annotation used to create training datasets.
A retail company is analyzing customer purchase data to develop personalized product recommendations. The company wants to use Amazon SageMaker Clarify to assess fairness metrics across different customer groups to avoid potential bias in the recommendation system.
The recommendation system needs to identify if certain customer segments are underrepresented in the training data. The company needs to choose a pre-training bias metric in SageMaker Clarify.
Which metric meets these requirements?
The correct answer is C. Class imbalance ratio.
Amazon SageMaker Clarify provides both pre-training and post-training bias detection capabilities to help data scientists ensure fairness in ML models. Pre-training bias metrics specifically analyze the training dataset to detect underrepresentation of certain groups or features before a model is trained. The class imbalance ratio measures how evenly different classes or sensitive attributes are represented in the dataset. A high imbalance indicates that a particular segment of customers (e.g., certain age groups, genders, or ethnicities) may be underrepresented, which could lead to biased predictions when the model is deployed.
Feature attribution bias (Option B) and prediction distribution skew (Option A) are post-training metrics. Feature attribution bias evaluates whether the model relies disproportionately on certain features for predictions, while prediction distribution skew examines whether the model's outputs are skewed for specific groups. Model performance gap (Option D) is also a post-training measure, focusing on differences in accuracy or other performance metrics between groups.
By using the class imbalance ratio during pre-training analysis, the company can identify and mitigate underrepresentation before training the model. Actions may include resampling the dataset, weighting underrepresented classes, or applying synthetic data augmentation to ensure that the model is trained on a balanced, fair dataset. This aligns with AWS best practices for ML solution monitoring, maintenance, and security, particularly for responsible AI and fairness assessment.
In summary, the class imbalance ratio allows proactive bias detection at the dataset level, preventing unfair recommendations and ensuring equitable treatment across all customer segments in the recommendation system.
An ML engineer receives datasets that contain missing values, duplicates, and extreme outliers. The ML engineer must consolidate these datasets into a single data frame and must prepare the data for ML.
Which solution will meet these requirements?
Amazon SageMaker Data Wrangler provides a comprehensive solution for importing, consolidating, and preparing datasets for ML. It offers tools to handle missing values, duplicates, and outliers through its built-in cleansing and enrichment functionalities, allowing the ML engineer to efficiently prepare the data in a single environment with minimal manual effort.
Exam domains verified against: Official Amazon MLA-C01 exam guide, last checked October 2026.
Ingest and store data using AWS data sources like S3 and streaming services, then transform and engineer features through cleaning, encoding, and scaling techniques. Use SageMaker Data Wrangler and AWS Glue for data exploration and ensure data integrity by addressing class imbalance and applying anonymization and compliance controls.
Select appropriate algorithms and AWS AI services for business problems, then train and refine models using techniques like distributed training, early stopping, and hyperparameter tuning. Analyze model performance using metrics such as confusion matrices, F1 scores, ROC curves, and identify overfitting and convergence issues.
Monitor model inference for drift using techniques to track data quality and model performance, then optimize infrastructure costs using CloudWatch and X-Ray. Secure ML resources through IAM roles and policies, apply SageMaker security features, and implement network access controls and CI/CD pipeline security.
Select deployment infrastructure by choosing between real-time, batch, and serverless endpoints, then create and script infrastructure using SageMaker, CloudFormation, or AWS CDK. Set up CI/CD pipelines with AWS CodePipeline and CodeBuild to automate model deployment with versioning and rollback strategies.
Common questions about the exam itself
