Task 2.3: Describe AWS infrastructure and technologies for building GenAI applications.

• AAmazon Quick

  Incorrect

  Amazon Quick is an AI-powered service for business users: chat, research, automation and analytics over connected data. It does not host or fine-tune models.

• BKiro

  Incorrect

  Kiro is an agentic coding service that turns prompts into specs, code and tests. It is not a model hosting platform.

• CAmazon SageMaker JumpStart

  Correct

  JumpStart provides pretrained, open-source models that you can fine-tune, evaluate and deploy, and SageMaker AI real-time endpoints run on the instance type you choose.

• DAmazon Comprehend

  Incorrect

  Amazon Comprehend is a managed NLP service with pre-trained models for entities, key phrases and sentiment. It does not deploy open-source foundation models.

Open model, own account, control over the instance: that is the SageMaker JumpStart path. Amazon Bedrock is the choice when you want no infrastructure to manage.

• AAmazon Bedrock Knowledge Bases

  Incorrect

  Knowledge Bases provide managed Retrieval Augmented Generation over your data. They do not host and run an agent built with another framework.

• BAmazon SageMaker JumpStart

  Incorrect

  JumpStart offers pretrained models and solution templates. It is not an agent runtime with memory and step tracing.

• CAmazon Lex

  Incorrect

  Amazon Lex builds conversational interfaces using voice and text. It does not run agents written with an open-source agent SDK.

• DAmazon Bedrock AgentCore

  Correct

  AgentCore works with frameworks such as Strands Agents and provides a serverless Runtime with session isolation, Memory for short-term and long-term context, and Observability to trace each step.

Frameworks such as Strands Agents define how an agent thinks; AgentCore provides what it needs in production: runtime, memory, identity, gateway and observability.

• AKiro

  Correct

  Kiro is an agentic coding service that turns prompts into specs and then into code, docs and tests, and its agent hooks run predefined actions on events such as saving files.

• BAmazon Quick

  Incorrect

  Amazon Quick serves business users with AI chat, research, flows, automation and dashboards over connected data. It is not a spec-driven coding tool for a code repository.

• CAWS Transform

  Incorrect

  AWS Transform uses agentic AI to migrate and modernize existing workloads, such as mainframe and VMware environments. It is not built for new feature work in a repository.

• DAmazon Bedrock Guardrails

  Incorrect

  Guardrails filter harmful content and protect model inputs and outputs. They do not write code.

Specs, code, tests and hooks in the developer's workspace describe Kiro; Quick is for business users and Transform for modernization projects.

• APurchase Provisioned Throughput for the model with a 6-month commitment

  Incorrect

  Provisioned Throughput is billed hourly for a fixed level of capacity, for as long as it is held, and batch inference is not supported for provisioned models.

• BRun the job as a batch inference job that reads from and writes to S3

  Correct

  Batch inference processes many prompts asynchronously from files in S3 and writes the responses back to S3, and select models are offered at a 50% lower price than on-demand inference.

• CSend every request with the Priority service tier

  Incorrect

  The Priority tier gives the fastest response times for a price premium over standard on-demand pricing. Speed is not needed overnight.

• DCall the model on demand one ticket at a time from a script

  Incorrect

  This works, but it pays full on-demand token prices for every ticket, which batch inference undercuts for this kind of job.

Large, latency-tolerant workloads with inputs already in S3 are what Bedrock batch inference is for, and it is priced below on-demand.

• AEvery response is guaranteed to be factually correct

  Incorrect

  Foundation models can hallucinate, producing incorrect or misleading output; no service removes that risk entirely.

• BThe company patches and scales the GPU servers behind each model

  Incorrect

  That is the work of self-hosting. With Bedrock, AWS runs the infrastructure and you call the models through an API.

• CThere is no infrastructure to manage, because the service is serverless

  Correct

  Amazon Bedrock is serverless, so you do not manage any infrastructure to use its foundation models.

• DAll prompts are sent to each model provider's own cloud to run

  Incorrect

  Your content is not shared with model providers, and you can keep traffic private between your VPC and Bedrock with AWS PrivateLink.

• ECustomer content is not used to improve the base models

  Correct

  With Amazon Bedrock, your content is not used to improve the base models and is not shared with any model providers.

Bedrock's pitch is a managed, serverless choice of models with your data kept private; it does not make model output infallible.

• AAmazon SageMaker JumpStart

  Incorrect

  JumpStart gives data scientists pretrained models to fine-tune and deploy. It is not a business user workspace.

• BAmazon Quick

  Correct

  Amazon Quick is an AI-powered service where users chat with agents over connected data and applications, with dashboards (Quick Sight), workflows (Quick Flows) and cited reports (Quick Research), fully managed.

• CKiro

  Incorrect

  Kiro is an agentic coding service for developers, turning prompts into specs, code and tests.

• DAWS Transform

  Incorrect

  AWS Transform uses agentic AI to migrate and modernize infrastructure, applications and code.

Amazon Quick is the business-user entry point to generative AI on AWS; Kiro serves developers, and Transform serves modernization projects.

• AAmazon Bedrock Knowledge Bases

  Incorrect

  Knowledge Bases retrieve information from documents for RAG. They do not expose APIs as callable tools.

• BAmazon Bedrock Prompt Management

  Incorrect

  Prompt Management stores and versions prompts; it does not connect agents to APIs.

• CAmazon Bedrock AgentCore Gateway

  Correct

  AgentCore Gateway converts APIs, Lambda functions and existing services into MCP-compatible tools and gives agents a single, secure endpoint to discover and use them.

• DAmazon SageMaker Feature Store

  Incorrect

  Feature Store stores ML features for training and inference. It has no role in agent tool access.

Gateway is the tool layer of AgentCore: wrap existing APIs once, and any MCP-compatible agent can use them.

• AThe Flex service tier

  Incorrect

  Flex offers a pricing discount for workloads that can handle longer processing times, the opposite of this requirement.

• BBatch inference

  Incorrect

  Batch inference processes prompts asynchronously from files in Amazon S3; it is not for interactive requests.

• CThe Standard service tier

  Incorrect

  Standard is the default tier with consistent performance for everyday tasks; Priority requests are served ahead of it.

• DThe Priority service tier

  Correct

  Priority delivers the fastest response times for a price premium over standard on-demand pricing, needs no prior reservation, and is prioritized over Standard and Flex requests.

Service tiers trade cost against speed: Flex for cheaper, slower work; Standard by default; Priority for a premium; Reserved for committed capacity.

• AFaster time to market, because it provides out-of-the-box RAG

  Correct

  Knowledge Bases abstract the heavy lifting of building pipelines and provide an out-of-the-box RAG solution, which reduces build time, and they remove the need to keep training the model on private data.

• BAnswers that are guaranteed to be free of any hallucinations or errors

  Incorrect

  RAG reduces hallucinations by grounding answers, but foundation models can still produce incorrect output; nothing guarantees zero errors.

• CNo need to choose an embeddings model or a vector store

  Incorrect

  When you set up a knowledge base you still choose an embeddings model and a vector store; quick-create can create the vector index for you.

• DA model that is permanently retrained on every new document

  Incorrect

  Knowledge Bases retrieve documents at query time precisely so the model does not have to be continually retrained.

Managed AWS generative AI services lower the barrier to entry: less to build, faster delivery and no training cycle for new data.

• AAmazon Lex

  Incorrect

  Amazon Lex is a managed service for building conversational interfaces with voice and text, not an open-source agent SDK.

• BStrands Agents

  Correct

  Strands Agents is an open-source SDK initially released by AWS for building agents with a model-first approach, with native MCP support and multi-agent patterns such as Swarm, Graph and Workflow.

• CAmazon SageMaker Canvas

  Incorrect

  SageMaker Canvas lets users generate ML predictions without writing code. It is not an agent framework.

• DAmazon Bedrock Prompt Management

  Incorrect

  Prompt Management stores and versions prompts; it is not a framework for writing agents.

Strands Agents is where an agent's logic is written; AgentCore is where such agents can run in production.