ML

A company wants to predict whether a student will pass an AWS certification examination. It has historical data containing years of IT experience, years of AWS experience, and time spent studying. Each historical record also contains the student's actual examination score.

The company wants to train a model and eventually predict the score of a new student.

Which workflow correctly describes this machine learning process?

Collect historical data, identify the label, train and tune a model, provide new data, and apply the trained model to generate a prediction.

A financial institution wants to predict whether a customer will default on a loan.

Each historical customer record contains:

• Annual income

• Credit score

• Employment history

• Existing debt

• Whether the customer actually defaulted

Which statement correctly identifies the features and label?

Annual income, credit score, employment history, and existing debt are features; default status is the label.

An e-commerce company develops several recommendation models. Each model uses customer purchase history, customer location, product preferences, and browsing activity.

The company wants to centrally organize, discover, securely access, and share these features across different models. It also wants consistency between model training and real-time predictions.

Which service best addresses this requirement?

Amazon SageMaker Feature Store

A data engineering team is organizing features in Amazon SageMaker Feature Store. Each feature group must identify individual records, describe their feature values, and record when the feature data was generated.

Which three components are associated with a Feature Group?

Record Identifier, Feature Name, Event Time

A financial analytics company needs to support both real-time predictions and historical model training.

It wants streaming access to recently ingested features and batch access to historical feature data.

Which architecture correctly supports these requirements?

Use the online store for streaming access through PutRecord/GetRecord APIs and the offline store in Amazon S3 for batch access.

A data scientist wants to query historical features stored in SageMaker Feature Store using Amazon Athena. They also want to prepare the historical data using SageMaker Data Wrangler.

Which statement is correct?

The offline store uses Amazon S3, supports tools such as Athena and Data Wrangler, and automatically creates a Glue Data Catalog.

A fintech company receives transaction events through Amazon Kinesis. It wants to engineer features from incoming transactions and make them available for real-time fraud predictions and periodic model retraining.

Which workflow best fits this requirement?

Send streaming feature records to SageMaker Feature Store's online store, use the online features for real-time inference, and use the offline S3 store for historical batch training.

An organization stores sensitive customer features in SageMaker Feature Store. It needs encryption at rest and in transit, fine-grained access control, and private connectivity.

Which combination aligns with the slides?

AWS KMS, IAM, and AWS PrivateLink

A pharmaceutical company must demonstrate exactly how its machine learning models were developed. Auditors need to trace data sources, processing jobs, training jobs, model history, and deployment activities.

Which SageMaker feature is most appropriate?

SafeMaker ML Lineage Tracking

A team is organizing its machine learning experiments in SageMaker.

Which statement correctly describes the relationship between trial components, trials, and experiments?

An experiment contains trials, and trials contain trial components such as processing, training, and transform jobs.

A data scientist is reviewing the other entities used by SageMaker ML Lineage Tracking.

Which matching is correct?

Context = logical grouping; Action = workflow step or model deployment; Artifact = object or data such as an S3 bucket or ECR image; Association = connection between entities.

A company is reviewing the relationships between its SageMaker lineage entities.

Which list contains the association types identified in your course?

ContributedTo, AssociatedWith, DerivedFrom, Produced, SameAs

An auditor wants to find every SageMaker model and endpoint that uses a particular S3 artifact. The team wants to retrieve lineage information programmatically through Python and generate a visual representation.

Which approach is correct?

Use the LineageQuery API through the SageMaker SDK for Python and an external Visualizer helper class for visualization.

A company is building an image classification model but does not have labeled training data. It needs people to identify what appears in thousands of images before the model can be trained.

Which service should it use?

Amazon SageMaker Ground Truth

A company sends 100,000 unlabeled images to Ground Truth. Initially, human labelers classify images. As the system learns, the model becomes more confident about certain classifications.

How does Ground Truth help reduce labeling costs?

It creates its own learning model and sends uncertain images to human labelers as the model improves.

An organization wants to create labeled training data using Ground Truth. It is considering the available human labeling resources.

Which options can it use?

Amazon Mechanical Turk, its own internal team, and professional labeling companies.

A company wants a turnkey data-labeling solution. It does not want to independently manage the labeling workflow or hire and coordinate the labeling workforce.

Which service best matches its requirement?

SageMaker Ground Truth Plus

A business analyst wants to build a predictive model without writing code. The analyst has a CSV dataset and wants to select a column to predict, build a model, and generate predictions.

Which service should the analyst use?

SageMaker Canvas

Which statement correctly describes SageMaker Canvas?

It supports classification and regression, automatic data cleaning, dataset joining, and sharing models and datasets with SageMaker Studio.

A data science team needs a visual interface to import, explore, transform, and prepare data before machine learning. It also wants to evaluate a model quickly using the prepared dataset.

Which service should it use?

SageMaker Data Wrangler

A team wants to import data into Data Wrangler from multiple AWS and external sources, then export the prepared data into SageMaker processing and pipeline workflows.

Which combination correctly describes its sources?

S3, Lake Formation, Athena, SageMaker Feature Store, Redshift, JDBC sources such as Databricks and SaaS.

A team wants to import data into Data Wrangler from multiple AWS and external sources, then export the prepared data into SageMaker processing and pipeline workflows.

Which combination correctly describes its targets?

SageMaker Processing, SageMaker Pipelines, and SageMaker Feature Store.

A data scientist is preparing a dataset using Data Wrangler. Which sequence correctly represents the main workflow shown in your slides?

Import Data → Preview Data → Visualize Data → Transform Data → Quick Model → Export Data Flow.

A data scientist receives an instance availability error while opening SageMaker Data Wrangler. Another team member cannot access a data source.

Which troubleshooting approach matches the course material?

Check Studio user IAM roles, data-source permissions, and EC2 instance quotas; request a quota increase if necessary.

A company wants to build a generative AI application. It is considering using a large, pretrained transformer model and adapting it to perform a specific task.

Which statement best describes a foundation model?

A large, pretrained model that can be fine-tuned for specific tasks or applied to new applications.

An organization wants to build a generative AI application using pretrained foundation models. It wants to invoke chat, text, and image models without provisioning and managing its own model-serving infrastructure.

Which service best addresses this requirement?

Amazon Bedrock

A developer attempts to access Amazon Bedrock but receives an authorization error. The developer has been using the AWS root account and has not configured the necessary permissions for the application user.

Which approach aligns with the course slides?

Use an appropriately permissioned IAM user and grant the relevant Bedrock permissions, such as AmazonBedrockFullAccess or AmazonBedrockReadOnly.

A developer wants to experiment with different foundation models and test chat, text, and image generation before integrating a model into an application.

Which Amazon Bedrock capability should the developer use?

Bedrock Playground

A company wants employees to ask questions about internal documents using a foundation model. The company wants the model to use current company information instead of relying solely on information learned during pretraining.

Which approach best meets the requirement?

Use Retrieval-Augmented Generation to retrieve relevant information from an external data source and incorporate it into the model's prompt.

A company is developing a RAG application. It needs to retrieve relationships between products, perform traditional keyword searches, and conduct semantic similarity searches.

Which statement is correct?

The database should be selected according to the type of information and retrieval required.

Graph databases can support relationships, traditional search engines can support keyword search, and vector databases support semantic similarity search.

A company wants to build a semantic search system. It needs to represent documents numerically so that documents with similar meanings are located close together in a multidimensional space.

What should the company generate?

Embeddings

An application receives a user search query. The application must find the most semantically similar documents from a database containing document embeddings.

Which sequence correctly describes vector retrieval?

Convert the query into an embedding vector, search the vector database for nearby vectors, and return the top-N most similar results.

A company wants to implement a document-question-answering application. It wants to upload its own documents, generate embeddings, store them in a vector database, and retrieve relevant information for an LLM.

Which configuration best represents Amazon Bedrock Knowledge Bases?

Data source + embedding model + vector store + Bedrock model for retrieval and generation.

An employee wants to ask questions about a company document. Another application wants to incorporate retrieved company information into an LLM agent.

Which statement best describes how Bedrock Knowledge Bases can be used?

They support automatic RAG for document conversations, integration into applications, and integration with agents through agentic RAG.

Pre-retrieval

indexing, chunking, data extraction, query rewriting

Retrieval

Retrieve relevant data

Post-retrieval

Augment the prompt and generate the response

A company is configuring a Bedrock Knowledge Base. It wants documents split into fixed-size chunks with a specified number of tokens and an overlap percentage.

Which chunking method should it use?

Standard fixed-size chunking

A developer configures standard chunking with a chunk size of 5 tokens and an overlap of 20%.

Assuming one token equals one word, how many words overlap between adjacent chunks?

1 word

A company wants precise retrieval from small embeddings but also wants to preserve more surrounding context when providing information to the LLM.

Which chunking approach should it use?

Hierarchical chunking

A developer configures hierarchical chunking with:

• Parent chunk size: 6

• Child chunk size: 3

• Overlap: 1

Assuming one token equals one word, which interpretation is correct?

Each parent chunk has a size of six tokens, smaller child chunks have a size of three tokens, and adjacent child chunks overlap by one token.

A company has documents containing closely related ideas. It wants chunks to be separated based on meaning rather than simply using a fixed number of tokens or sentences.

Which chunking method should it select?

Semantic chunking

A developer configures semantic chunking in Amazon Bedrock.

Which parameters should the developer expect to configure?

Maximum tokens, buffer size, and breakpoint percentile threshold.

A company configures semantic chunking. It wants to understand the effects of buffer size and the breakpoint percentile threshold.

Which statement is correct?

A buffer size of 1 considers three sentences.

Too large a buffer can introduce noise, while too small a buffer can miss important context but improve precision.

A higher breakpoint threshold produces fewer, larger, more distinguishable chunks.

A company has a large vector database and wants to reduce storage and processing costs. It is considering reducing the number of dimensions in its embedding vectors.

Which statement correctly describes the tradeoff?

Smaller vector dimensions can reduce cost, but may reduce retrieval performance. The correct dimensionality depends on the complexity of the information being searched.

A data engineering team is comparing sparse and dense embeddings for a semantic search application.

Which statement is correct?

Sparse embeddings are large vectors containing mostly empty values; dense embeddings are more compact and encode semantic information more efficiently.

A company wants to determine how similar two embedding vectors are. It wants to measure their directional similarity using the angle between them.

Which similarity measure is commonly used?

Cosine similarity

A company stores document chunks and embeddings in a Bedrock Knowledge Base. It also wants to filter retrieval based on document category, access control, topic, document ID, and section.

Which approach is most appropriate?

Store useful metadata alongside the chunks and embeddings, and use metadata filtering to improve retrieval and relevance.

An organization stores its source documents in Amazon S3. When a document changes, the organization wants to generate updated embeddings and synchronize its Bedrock Knowledge Base.

Which workflow best matches the course material?

An S3 event triggers AWS Lambda, which can generate updated embeddings and start a Knowledge Base ingestion job. The process can also be scheduled for batch updates.

A company is building a Bedrock Knowledge Base and needs a vector store to hold document embeddings and support similarity searches.

Which service does your course emphasize as the primary implementation for backing Bedrock Knowledge Bases?

Amazon OpenSearch Service

An organization wants its search application to understand the meaning of a user's query while also considering exact keyword matches.

Which search approach should it use?

Hybrid search

A data engineering team is configuring an OpenSearch vector index and wants to understand the available top-level vector engines.

Which list is correct?

Facebook AI Similarity (FAISS), Non-Metric Space Library (NMSLib), and Apache Lucene.

A company has a large vector database. It wants to retrieve similar vectors but needs to balance retrieval accuracy, speed, and memory usage.

Which statement is correct?

Exact Nearest Neighbor searches are slower, while Approximate Nearest Neighbor (ANN) methods improve search efficiency by avoiding exhaustive exact comparisons.

An organization is choosing an Approximate Nearest Neighbor method for OpenSearch.

It has two different requirements:

• Workload A: Fast, high-quality vector searches, with sufficient memory available.

• Workload B: Extremely large datasets where reducing memory usage and improving search speed may require sacrificing some recall.

Which pairing is correct?

Workload A = Hierarchical Navigable Small World (HNSW); Workload B = Inverted File (IVF)

An engineer is tuning an HNSW vector index. The engineer increases the M parameter to improve the number of connections between nodes.

What is the expected effect?

The graph becomes denser, potentially increasing recall but requiring more memory.

An engineer wants to improve the accuracy of the HNSW graph when building the index. The engineer is willing to accept slower indexing.

Which parameter should be increased?

ef_construction

A company has already built its HNSW index. It now wants to improve recall during searches, accepting that individual searches may take longer.

Which parameter should it increase?

ef_search

M

Edges per node

More memory, potentially higher recall

ef_construction

Graph construction candidate list

More accurate graph, slower indexing

ef_search

Search exploration

Higher recall, slower searching

A company has a large vector-search workload and wants to reduce the storage required for its vector representations. It is considering binary vectors.

Which statement correctly describes binary vectors in your course?

They represent bit sequences stored as bytes and provide fast distance searching, with approximately 32× compression compared with float32.

A company wants to reduce embedding storage requirements without converting the vectors into binary bit sequences.

It decides to use FP16.

What happens?

Each 32-bit floating-point value is represented using a 16-bit floating-point value.

A company is configuring OpenSearch shards for two different workloads:

• Workload A: Primarily semantic vector search.

• Workload B: Hybrid search combining vector and keyword search.

Which configuration matches the recommendations in your slides?

Semantic: 30–50 GB shards; Hybrid: 10–30 GB shards.

A company maintains several different document collections. Each collection has different data characteristics and may require different embedding models and performance configurations.

Which OpenSearch strategy is most appropriate?

Create separate indexes for different document types, tune each index for its corpus, and use query routing to retrieve from the appropriate indexes.

A company already uses Amazon Aurora PostgreSQL for structured business data. It wants to add vector similarity searches while keeping structured and vector data in the same database.

Which option is correct?

Use the pgvector extension

A company is building a small-to-medium RAG application with mostly structured data. It requires complex filtering logic and wants to store both structured and vector data together.

Which option may be appropriate?

Amazon Aurora with pgvector

A company wants to provide employees with an AI assistant that can answer questions about company documents, summarize information, generate content, and automate routine tasks.

Which service should it use?

Amazon Q Business

A company wants Amazon Q Business to access information from Microsoft 365, Salesforce, SharePoint, and Slack. It also wants employees to interact with third-party services such as Jira and ServiceNow.

Which statement correctly distinguishes connectors and plugins?

Data connectors retrieve and index information from enterprise sources, while plugins allow interaction with third-party services and execution of actions.

An organization uses Amazon Q Business to answer employee questions. It wants employees to receive answers only from documents they are authorized to access.

Which configuration best addresses this requirement?

Integrate Amazon Q Business with IAM Identity Center so users are authenticated and responses respect their document permissions.

A company wants to prevent its Amazon Q Business assistant from discussing certain topics. It also wants responses to use only internal company information rather than external knowledge.

Which feature should administrators configure?

Amazon Q Business administrative controls and guardrails.

A business analyst wants to create a generative AI-powered application using natural language. The analyst does not know how to code and wants the application to use internal company information and potentially interact with third-party services.

Which service should the analyst use?

Amazon Q Apps

An organization has thousands of documents across S3, RDS, Google Drive, and SharePoint. Employees need to search these documents using natural language and receive answers extracted from within the documents.

Which service best fits this requirement?

Amazon Kendra

An organization uses Amazon Kendra for employee searches. It notices that employees frequently select certain search results over others.

The organization wants Kendra to learn from user interactions and feedback to improve the relevance of future search results.

Which capability should it use?

Incremental Learning

An organization wants to manually adjust search result rankings so that more important, fresher, or customized information appears higher in search results.

Which service capability addresses this requirement?

Amazon Kendra relevance tuning

A company has two separate requirements:

• Requirement A: Employees need an intelligent document search service that can extract answers from PDFs, PowerPoint files, Word documents, and FAQs.

• Requirement B: A developer wants to build a generative AI assistant that retrieves relevant internal documents and supplies them to a foundation model to generate responses.

Requirement A = Amazon Kendra; Requirement B = Amazon Bedrock Knowledge Bases.

A company needs labeled training data but wants to automate some of the labeling process rather than relying exclusively on human labelers.

Which AWS service pair matches the alternatives described in your course?

Amazon Rekognition for image recognition and Amazon Comprehend for text analysis.

A financial analytics company has:

• Historical stock data stored in Amazon RDS.

• Real-time stock market events arriving through Amazon Kinesis.

• A requirement to prepare features, train a predictive model, and generate predictions.

Which workflow best describes the intended architecture?

Ingest historical and streaming data, prepare and engineer features, use SageMaker to train and tune a model, deploy the model, and apply it to incoming data to generate predictions.