Sagemaker Bedrock

What does Amazon Bedrock provide?
A) An API for managing EC2 instances
B) An API for generative AI Foundation Models
C) An API for building mobile apps
D) An API for IoT devices

Which of the following is true about models in Amazon Bedrock?
A) You can only use AWS pre-built models
B) You can use only third-party models
C) You can use pre-built, fine-tuned, or your own models
D) You must train all models yourself

Which Bedrock API endpoint is used for inference against models?
A) bedrock
B) bedrock-runtime
C) bedrock-agent
D) bedrock-agent-runtime

Which Bedrock API endpoint is used for inference against agents and knowledge bases?
A) bedrock-runtime
B) bedrock
C) bedrock-agent
D) bedrock-agent-runtime

Who has access to Bedrock?
A) Must use with an IAM user
B) Root user
C) Root user and IAM users

What must you do before using any base model (ex: Amazon Titan) in Amazon Bedrock?
A) Deploy the model manually
B) Request access to the model
C) Upload the model to S3
D) Pay an upfront fee

What does the Bedrock “playground” environment allow you to do?
A) Test chat, text, and image models
B) Only deploy models to production
C) Tune the AWS network settings
D) Train new models from scratch

What does Semantic Chunking do in Bedrock knowledge bases?
A) Splits data by fixed size intervals
B) Uses a foundation model to group semantically related sentences
C) Groups data based on predefined metadata
D) Partitions documents by timestamp

What is Hierarchical Chunking in Bedrock knowledge bases?
A) Groups random smaller chunks together
B) Creates parent chunks with smaller child chunks
C) Aggregates chunks by keyword frequency
D) Segments data based on document length

What is the purpose of the Lambda-based chunking strategy in Bedrock?
A) Execute chunking based on model-defined templates
B) Allow users to define chunking logic via Lambda functions
C) Perform chunking at the metadata layer
D) Automatically chunk based on document schema

What do Amazon Bedrock Guardrails primarily filter?
A) Training datasets
B) Prompts and model responses
C) S3 input data
D) API request payloads

Which type of models do Bedrock Guardrails support?
A) Vision models
B) Audio models
C) Text foundation models
D) Tabular models

What does the Contextual Grounding Check in Bedrock Guardrails help prevent?
A) Data leakage
B) Hallucination
C) Model drift
D) Overfitting

Where can Amazon Bedrock Guardrails be incorporated?
A) Only into custom agents
B) Only into knowledge bases
C) Agents and knowledge bases
D) SageMaker pipelines

What can be customized when Bedrock Guardrails block a message?
A) The token usage limit
B) The blocked message response
C) The model timeout configuration
D) The Lambda trigger response

What does Amazon Bedrock provide?
A) An API for deploying infrastructure
B) An API for generative AI foundation models
C) An API for managing SageMaker endpoints
D) An API for orchestrating serverless workflows

Which Bedrock API endpoint is used for inference against foundation models?
A) bedrock
B) bedrock-runtime
C) bedrock-agent
D) bedrock-agent-runtime

Which Bedrock API endpoint is used for inference against agents and knowledge bases?
A) bedrock-runtime
B) bedrock
C) bedrock-agent
D) bedrock-agent-runtime

What must you do before using a model in Amazon Bedrock?
A) Request model access
B) Complete a fine-tuning process
C) Deploy a training job
D) Attach an IAM policy

Which networking service is recommended when using sensitive training data for fine-tuning in Bedrock?
A) AWS Global Accelerator
B) Amazon VPC with PrivateLink
C) AWS Direct Connect
D) Amazon Route 53

What is the purpose of Continued Pre-Training in model development?
A) Fine-tuning a model with labeled prompt-completion pairs
B) Training a model from scratch using proprietary data
C) Exposing a model to additional unlabeled text to enhance familiarity with domain-specific content
D) Reducing the model’s inference latency by optimizing token usage

What is a primary advantage of using Retrieval-Augmented Generation (RAG) compared to fine-tuning?
A) Faster and cheaper incorporation of new or proprietary information
B) Increased model parameter size
C) Improved model latency without data updates
D) Directly embedding new knowledge into the model weights

In a RAG architecture, how is information typically retrieved?
A) Directly from model weights
B) From a relational database
C) Using semantic search over a vector store
D) By polling multiple APIs

Which of the following is a potential disadvantage of RAG-based systems?
A) They cannot hallucinate
B) They are deterministic in their outputs
C) They are highly sensitive to prompt template design
D) They cannot incorporate new information after deployment

Which type of database is most commonly used for RAG (Retrieval-Augmented Generation) architectures?
A) Relational database
B) Graph database
C) Vector database
D) Time series database

Which database types can function as a vector database for RAG?
A) Only purpose-built vector databases like Pinecone
B) Only relational databases like PostgreSQL
C) OpenSearch or Elasticsearch with vector support
D) Only graph databases like Neo4j

What is the correct order of operations for performing retrieval-augmented generation (RAG)?

When creating a Bedrock Knowledge Base, what does controlling the “chunking” process allow you to manage?
A) The rate of document ingestion
B) The size of each embedding vector
C) The number of tokens each vector represents
D) The number of concurrent API calls

Describe the design of RAG in Bedrock

You are configuring a Bedrock agent and want to ensure it retrieves the most accurate information for user queries. Which TWO actions would most improve retrieval quality? (Select TWO)
A) Decrease the chunk size when ingesting documents into the Knowledge Base
B) Fine-tune the embedding model on your specific domain
C) Use Guardrails to automatically re-rank search results
D) Use a higher-dimension vector space for embeddings
E) Increase the Provisioned Throughput

Name 2 data design considerations in RAG Knowledge Bases

What is the purpose of the Planning Module in a Bedrock LLM agent?
A) Execute external Lambda functions
B) Store user authentication credentials
C) Guide the LLM in breaking down a user request into sub-questions
D) Schedule model retraining

What is the function of the Tools Core in Bedrock LLM agents?
A) Provide a set of Lambda functions or APIs the agent can invoke
B) Manage model versioning
C) Translate user inputs into embeddings
D) Configure security groups for external services

In Bedrock LLM agents, what is considered the agent’s "memory"?
A) Only embeddings generated during a session
B) External S3 storage of all user data
C) A combination of chat history and external data sources
D) Fine-tuned model parameters

How does a Bedrock LLM agent decide which tool to use in response to a user request?
A) Manual assignment at runtime
B) Selection based on Action Groups and associated prompts that describe the tools
C) Predefined tool selection configured in the agent’s IAM policy
D) Randomized tool invocation for exploration

What is the primary function of the optional Code Interpreter in Bedrock LLM agents?
A) Automatically generates optimized prompts for external tools
B) Encrypts user inputs before external API calls
C) Allows the agent to dynamically write and execute code to assist with user tasks
D) Fine-tunes the foundational model during extended conversations to improve output quality

What is the purpose of creating an "Alias" when deploying an LLM agent in Amazon Bedrock?
A) It grants the agent permission to access external APIs
B) It creates a deployed snapshot of the agent for versioned access
C) It automatically increases the agent’s throughput limits
D) It encrypts all memory associated with the agent

What is the key difference between On-Demand Throughput (ODT) and Provisioned Throughput (PT) in Bedrock agent deployment?
A) ODT allows scaling beyond account quotas, while PT restricts usage to a fixed number
B) ODT uses purchased token capacity, while PT uses quotas set at the account level
C) ODT relies on account-level quotas without pre-purchased capacity, while PT allows purchasing increased token rates
D) ODT encrypts agent responses by default, while PT does not

A user deployed an agent in Amazon Bedrock using an alias. However, they notice throttling errors during periods of high traffic. What is the most likely cause?
A) The alias was not correctly mapped to the agent’s memory
B) The agent is using On-Demand Throughput and has hit the account-level quota limits
C) The alias was not encrypted with KMS
D) The agent requires an updated embedding model for retrieval

A company fine-tuned a Titan model in Bedrock to better handle internal customer support queries. After deployment, they notice that newer FAQs are still answered incorrectly. What is the most efficient way to update the model's knowledge without initiating another fine-tuning process?
A) Retrain the fine-tuned model with updated prompts
B) Create a new Bedrock Knowledge Base and integrate it using RAG
C) Enable automatic fine-tuning updates on the model endpoint
D) Adjust the model’s hyperparameters to increase relevance

A Bedrock agent is integrated with several Lambda tools. During testing, the agent often fails to invoke the correct tool, even though all tools are properly deployed. What is the most likely cause of this issue?
A) The tools were not registered using Bedrock Action Groups with descriptive prompts
B) The Lambda functions are configured with too much memory
C) The agent alias is misconfigured
D) The agent is using Provisioned Throughput but has insufficient quotas

A user configured Bedrock Guardrails to prevent PII exposure in chat responses. However, during testing, some responses still leak email addresses. What should the user check first?
A) The model endpoint encryption settings
B) If the Guardrails are properly associated with both prompts and responses
C) If the model is using a provisioned vector store
D) If the Knowledge Base chunk size is too large

An engineer uploading long technical documents into a Bedrock Knowledge Base notices that search results often return irrelevant sections. What adjustment would most likely improve the relevance of retrieval?
A) Increase the token limit for the prompts
B) Decrease the chunk size during document ingestion
C) Use On-Demand Throughput instead of Provisioned Throughput
D) Add an external Lambda retriever

After deploying a Bedrock agent with Provisioned Throughput, a team notices that the system scales smoothly but costs are significantly higher than expected. What is the most likely explanation?
A) Provisioned Throughput incurs additional charges based on reserved token capacity
B) The alias deployment requires a premium support plan
C) The agent’s memory is causing increased vector storage costs
D) Guardrails filtering increases throughput token consumption

You are configuring a Bedrock agent and want to ensure it retrieves the most accurate information for user queries. Which TWO actions would most improve retrieval quality? (Select TWO)
A) Decrease the chunk size when ingesting documents into the Knowledge Base
B) Fine-tune the embedding model on your specific domain
C) Use Guardrails to automatically re-rank search results
D) Use a higher-dimension vector space for embeddings
E) Increase the Provisioned Throughput

You are deploying a Bedrock agent integrated with Lambda tools. To ensure the agent can correctly use these tools, which TWO configurations are essential? (Select TWO)
A) Define Action Groups with descriptive prompts
B) Fine-tune the foundational model with tool instructions
C) Configure tool input parameters including name, description, and type
D) Enable Guardrails for input validation
E) Set up On-Demand Throughput before tool registration

How does an application interact with a deployed Bedrock agent at runtime?
A) By calling the model endpoint directly with a prompt
B) By using the InvokeAgent request with the agent’s alias ID and the Agents for Amazon Bedrock Runtime Endpoint
C) By invoking a Lambda function registered in the agent’s Action Group
D) By submitting a batch inference job to the Bedrock runtime

What are two supported sources for importing models into Amazon Bedrock?
A) AWS Lambda and Amazon S3
B) Amazon S3 and SageMaker
C) Amazon RDS and OpenSearch
D) SageMaker and DynamoDB

Which evaluation metrics can be automatically assessed when testing a model in Amazon Bedrock?
A) Inference latency and cost only
B) Accuracy, toxicity, robustness, BERTScore, F1
C) BLEU Score and TF-IDF Score only
D) Storage efficiency and throughput

In Amazon Bedrock, who can perform human evaluation of model outputs?
A) Only AWS employees
B) Only Bedrock system administrators
C) Either your own team or an AWS-managed team
D) Only IAM users with Administrator permissions

What does Bedrock’s watermark detection feature identify?
A) Whether an image has been encrypted with KMS
B) Whether an image was generated by the Titan model
C) Whether an image was uploaded from an external database
D) Whether an image exceeds the maximum token limit

What is Amazon Bedrock Studio?
A) A serverless API gateway for model endpoints
B) A no-code platform for training foundation models
C) A web-based workspace for Bedrock without AWS accounts, using SSO integration
D) A CLI tool for managing Bedrock model deployments