AWS Learning Material Certification Cheatsheets
AIF-C01 Domain 2 24% domain

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

Fundamentals of GenAI · 4,708 words · source: Certified-AI-Practitioner-AIF-C01/domain-2/task-2-3-cheat-sheet.md

Domain 2: Fundamentals of GenAI

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

Domain 2 is 24% of scored exam content. Task 2.3 focuses on AWS services for building GenAI applications, why managed AWS GenAI services reduce implementation burden, how AWS infrastructure supports security and reliability, and how cost changes with model choice, tokens, throughput, region, and customization.

Use this sheet for service-matching and tradeoff questions. The exam often asks which AWS service fits a GenAI use case, why managed services are useful, which AWS infrastructure benefit matters, or what cost tradeoff explains a design choice.

Exam Objective Map

You should be able to:

AWS exam objective What to know for the exam
Identify AWS services and features to develop GenAI applications Amazon Bedrock, Amazon SageMaker AI, SageMaker JumpStart, Amazon Quick, Amazon Q, Kiro, Strands Agents, Amazon Bedrock AgentCore
Describe advantages of AWS GenAI services Accessibility, lower barrier to entry, efficiency, cost-effectiveness, faster time to market, business-objective alignment
Describe AWS infrastructure benefits Security, compliance, shared responsibility, safety, availability, fault tolerance, global Regions and Availability Zones
Describe cost tradeoffs Token-based pricing, model/provider/modality pricing, latency, availability, redundancy, region coverage, Provisioned Throughput, custom models, self-hosted infrastructure

The Big Picture

AWS gives you multiple layers for GenAI workloads:

Application layer
  Amazon Q, Amazon Quick, Kiro, custom apps, assistants, dashboards, workflows

GenAI platform layer
  Amazon Bedrock, Bedrock Knowledge Bases, Bedrock Agents, Guardrails, AgentCore,
  SageMaker AI, SageMaker JumpStart, Strands Agents

Infrastructure layer
  AWS Regions, Availability Zones, VPC, IAM, KMS, CloudTrail, CloudWatch,
  Nitro, GPUs, AWS Trainium, AWS Inferentia, storage, networking

Exam shortcut:

Need managed FM API -> Amazon Bedrock.
Need custom ML/FM build, train, tune, deploy -> Amazon SageMaker AI.
Need pretrained model hub in SageMaker -> SageMaker JumpStart.
Need enterprise assistant over company data -> Amazon Q Business.
Need developer assistant -> Amazon Q Developer, Kiro.
Need BI with natural-language insights -> Amazon Quick.
Need production agents with identity, memory, tools, runtime, observability -> Bedrock AgentCore.
Need open-source agent SDK -> Strands Agents.

AWS GenAI Service Chooser

Need Best AWS anchor Exam-ready meaning
Access foundation models through an API without managing model infrastructure Amazon Bedrock Fully managed service for using FMs from Amazon and third-party providers.
Build and scale GenAI apps with RAG, agents, guardrails, and model evaluation Amazon Bedrock features Bedrock is the main managed GenAI application platform.
Experiment with model responses and inference parameters Amazon Bedrock playgrounds Try prompts and compare model behavior before application integration.
Add enterprise or private knowledge to model responses Amazon Bedrock Knowledge Bases, Amazon Q Business Managed RAG-style grounding with retrieved data.
Add safety and privacy controls to prompts and responses Guardrails for Amazon Bedrock Content filters, denied topics, word filters, sensitive information filters, contextual grounding, and automated reasoning checks.
Build agents that call tools or APIs Amazon Bedrock Agents, Amazon Bedrock AgentCore, Strands Agents Agents use FMs plus tools, orchestration, memory, and permissions.
Deploy and operate production agents securely at scale Amazon Bedrock AgentCore Agent runtime, memory, gateway, identity, code interpreter, browser, observability, policy, registry, and evaluations.
Build autonomous agents with an open-source SDK Strands Agents Model-first SDK with AWS integrations, tool support, MCP support, and multi-agent patterns.
Build, train, tune, evaluate, or deploy ML/custom models Amazon SageMaker AI More control over ML lifecycle and custom model workflows than a pure managed FM API.
Start from pretrained FMs and example notebooks in SageMaker SageMaker JumpStart Model hub for pretrained and foundation models, solution templates, examples, fine-tuning, and deployment.
Help employees answer questions, summarize, generate content, or complete tasks using enterprise data Amazon Q Business Managed enterprise GenAI assistant with connectors, permissions-aware responses, and citations.
Help developers write, understand, secure, and modernize code Amazon Q Developer GenAI assistant for AWS building, IDE coding help, code generation, security scanning, debugging, and upgrades.
Turn prompts into specs, code, docs, and tests Kiro Agentic coding service / IDE for spec-driven development, steering files, hooks, and implementation plans.
Ask natural-language questions of business data and generate BI narratives Amazon Quick Generative BI capabilities such as Q&A, data stories, executive summaries, natural-language visual and calculation authoring.

Exam trap: Amazon Bedrock and SageMaker AI can both be used with foundation models, but they are not the same.

Bedrock = managed FM access and GenAI application features.
SageMaker AI = broader ML platform for building, training, tuning, hosting, and operating models.
JumpStart = SageMaker model hub and starter assets.

Amazon Bedrock: What To Associate

Amazon Bedrock is the central AWS service for managed foundation-model access and GenAI application building.

Bedrock capability What it helps with Exam signal phrase
Model catalog / model access Choose among supported FMs from Amazon and third-party providers. "Use different foundation models by API"
Runtime APIs Invoke FMs from applications. "Call a model without hosting it"
Playgrounds Experiment with prompts, model outputs, and parameters. "Test prompt behavior before building"
Knowledge Bases Managed RAG over enterprise data. "Answer from company documents"
Agents Let an FM orchestrate steps and call tools/APIs. "Take action, call Lambda, query systems"
Guardrails Apply safety, privacy, and policy controls. "Block harmful content or mask PII"
Model evaluation Compare models and evaluate outputs. "Select best model for use case"
Model customization Fine-tune or customize supported models. "Adapt model behavior to domain or task"
Prompt management Reuse and version prompts. "Manage prompt templates"

Exam shortcut:

Bedrock = models + APIs + RAG + agents + guardrails + evaluation.

SageMaker AI and JumpStart

Use SageMaker AI when the scenario is about the ML lifecycle, custom model development, or more control over training and hosting.

Service / feature Use it for Cost/operations note
Amazon SageMaker AI Build, train, tune, deploy, monitor, and operate ML models. You manage more lifecycle choices than with a fully managed FM API.
SageMaker JumpStart Discover pretrained models, foundation models, examples, notebooks, and solution templates. Useful for quick starts, fine-tuning, and deploying models into SageMaker workflows.
SageMaker endpoints Host models for inference. Running endpoints can continue to incur cost; delete unused endpoints.
SageMaker training/fine-tuning Train or adapt models with data. GPU/accelerator compute and storage can be significant.

Skill Builder memory hook:

Most teams will not train an LLM from scratch.
Use pretrained models, transfer learning, managed APIs, or model hubs to reduce time and cost.

Exam traps:

  • JumpStart is not the same as Bedrock. JumpStart is inside the SageMaker ecosystem.
  • SageMaker AI is broader than GenAI. It supports traditional ML and custom model lifecycle work.
  • Training from scratch is usually the most expensive and complex option.
  • Fine-tuning and hosting can require expensive accelerator resources.

Amazon Q, Amazon Quick, and Kiro

These are application-layer or productivity-layer GenAI services.

Service Best association Exam signal phrase
Amazon Q Business Enterprise assistant over business data. "Employees ask questions from internal documents with citations and permissions"
Amazon Q Developer Developer and AWS-building assistant. "Generate code, explain code, scan for vulnerabilities, answer AWS questions"
Amazon Quick Generative BI and natural-language analytics. "Ask questions of business data, generate dashboards, data stories, executive summaries"
Kiro Agentic coding and spec-driven development. "Turn prompt into specs, implementation plan, code, tests, and docs"

Exam shortcut:

Q Business = enterprise knowledge and tasks.
Q Developer = developers and AWS workloads.
Quick = BI and analytics insights.
Kiro = agentic coding workflow.

Strands Agents vs Amazon Bedrock AgentCore

Concept What it is Best association
Strands Agents Open-source SDK initially released by AWS for building autonomous AI agents. Developer framework for model-first agents, tools, MCP, AWS integrations, multi-agent patterns
Amazon Bedrock AgentCore Managed platform for deploying and operating agents securely at scale using any framework and model. Production agent runtime, memory, gateway, identity, policy, observability, evaluation

Simple distinction:

Strands Agents helps build agents.
AgentCore helps run, secure, govern, connect, observe, and scale agents.

AgentCore capability map:

AgentCore capability What to remember
Runtime Secure serverless hosting for agents and tools.
Memory Short-term and long-term context for agents.
Gateway Converts APIs, Lambda functions, and services into MCP-compatible tools.
Identity Agent identity, access, and authentication management.
Code Interpreter Isolated sandbox for code execution and analysis.
Browser Managed browser runtime for web interaction.
Observability Trace, debug, monitor, and audit agent behavior.
Policy Deterministic control over which tools/actions agents can use.
Registry Catalog for agents, tools, MCP servers, skills, and resources.
Evaluations Assess agent and tool behavior before and after deployment.

Exam trap: Agents create new security and cost questions because they can call tools, store memory, and take actions. Use least privilege, identity, policy, monitoring, validation, and guardrails.

Advantages of AWS GenAI Services

Advantage What it means Exam signal phrase
Accessibility Teams can use models and GenAI capabilities through managed services, APIs, consoles, and SDKs. "No deep ML expertise required"
Lower barrier to entry Avoid building or training FMs from scratch. "Start from managed FMs or pretrained models"
Efficiency Use managed infrastructure, model hubs, templates, APIs, and built-in integrations. "Reduce undifferentiated heavy lifting"
Cost-effectiveness Pay for use, choose right-sized models, avoid idle self-hosted GPUs, use managed scaling. "Do not buy infrastructure for variable workloads"
Speed to market Quickly prototype, evaluate, deploy, and iterate. "Launch faster with managed services"
Business alignment Choose services by business outcome: support, productivity, BI, coding, automation, customer experience. "Meet business objectives"
Security and privacy integration IAM, encryption, VPC/PrivateLink patterns, logging, monitoring, guardrails, and compliance documentation. "Enterprise-grade security controls"
Scalability and availability Use AWS global infrastructure and managed service resilience. "Scale to many users or requests"

Memory hook:

AWS GenAI services reduce model-building burden, infrastructure burden, and operations burden.

AWS Infrastructure Benefits for GenAI

Infrastructure benefit Why it matters for GenAI AWS anchors
Regions Place workloads near users and meet data residency or regional availability requirements. AWS Regions, model/service regional coverage
Availability Zones Design for high availability and fault tolerance. Multi-AZ architecture, redundant networking
Security of the cloud AWS protects the infrastructure that runs AWS services. Shared Responsibility Model
Security in the cloud Customer controls data, identity, application behavior, permissions, and service configuration. IAM, KMS, VPC, CloudTrail, CloudWatch
Compliance support AWS provides reports and attestations; customer still owns workload compliance decisions. AWS Artifact, AWS Audit Manager, AWS Config
Data protection Encrypt data, restrict access, log activity, avoid sensitive data in tags/free-form fields. AWS KMS, IAM, CloudTrail, Macie
Private connectivity Keep traffic private where supported. Amazon VPC, AWS PrivateLink
Safety controls Detect/filter unsafe content, PII, prompt attacks, hallucination risk, and policy violations. Guardrails for Amazon Bedrock
Specialized compute Improve price/performance for AI training or inference. AWS Trainium, AWS Inferentia, GPUs, Nitro-based instances
Observability Monitor usage, token metrics, latency, errors, traces, and agent behavior. CloudWatch, CloudTrail, AgentCore Observability

Shared responsibility shortcut:

AWS secures the cloud infrastructure.
You secure your data, access, prompts, retrieved content, application logic, tool permissions, and outputs.

Exam traps:

  • Compliance is shared. AWS provides infrastructure controls and compliance artifacts, but the customer is responsible for workload-specific compliance.
  • Bedrock model providers do not get access to customer prompts and completions through Bedrock deployment accounts.
  • Guardrails reduce risk; they do not remove the need for application validation, monitoring, and human review in high-risk workflows.
  • Region choice affects latency, availability, data residency, service availability, and model availability.

The Three AI System Components To Protect

Skill Builder emphasizes three critical components:

Input -> Model -> Output
Component What can go wrong Controls
Input Prompt injection, sensitive data leakage, malicious instructions, poor-quality retrieved content Input validation, guardrails, access control, redaction, retrieval filters
Model Wrong model choice, unsafe behavior, bias, lack of grounding, model/version changes Model evaluation, versioning, responsible AI checks, monitoring, model selection
Output Hallucination, toxic content, PII exposure, unsafe tool instructions, wrong format Output filtering, grounding checks, structured validation, citations, human review

Cost Tradeoffs for AWS GenAI Services

Cost questions usually test tradeoffs, not exact prices.

Cost factor What increases cost What can reduce or control cost
Input tokens Long prompts, many examples, chat history, retrieved chunks, tool definitions Shorter prompts, better retrieval, summarization, prompt caching where supported
Output tokens Long answers, high max_tokens, verbose responses Output limits, concise prompts, task-specific response formats
Model choice Larger or more capable models, expensive providers, multimodal/image/video models Use right-sized model, route simple tasks to cheaper/faster models
Modality Images, video, audio, embeddings, and multimodal processing can have different pricing models Match model to actual modality need
Latency and responsiveness Low-latency requirements can require faster models, optimized tiers, or provisioned capacity Use streaming, smaller models, caching, async/batch for noninteractive work
Availability and redundancy Multi-Region, high availability, and reserved capacity can add cost Match resilience to business criticality
Regional coverage Model/service may not be available in every Region; region choice can change latency and price Choose Regions based on data residency, model availability, latency, and compliance
Provisioned Throughput Fixed hourly cost for higher/guaranteed throughput; billing continues until deleted Use for predictable high-volume or custom-model workloads, not low/spiky traffic
Custom models Training/fine-tuning, storage, evaluation, hosting, and provisioned capacity can add cost Prefer prompting/RAG first when the need is missing knowledge
SageMaker hosting Endpoints and accelerator instances can run continuously Delete unused endpoints, right-size instances, monitor utilization
Self-hosted LLMs Compute, GPUs/accelerators, storage, operations, scaling, licenses Use managed services when variable usage or low ops burden matters

Token cost shortcut:

Total GenAI request cost is driven by model/provider/modality plus input and output usage.
Longer context can improve quality but increases cost, latency, and quota pressure.

Provisioned Throughput shortcut:

On-demand = flexible, pay per use, good for variable workloads.
Provisioned Throughput = fixed hourly capacity, better for predictable scale or guaranteed throughput.
Custom model = may add training, storage, and dedicated inference capacity costs.

Exam traps:

  • Bigger model is not automatically the best answer; it may cost more and respond more slowly.
  • A longer prompt is not free; input tokens can matter as much as output tokens.
  • Setting max_tokens too high can reduce throughput/concurrency even if the final answer is shorter.
  • Provisioned Throughput can improve predictability but can waste money if traffic is low or spiky.
  • Batch or asynchronous processing can reduce cost or improve throughput when users do not need immediate responses.
  • Self-hosting gives control but shifts infrastructure, scaling, patching, monitoring, and utilization risk to you.

Managed Service vs Self-Hosted Model

Option Best for Tradeoff
Amazon Bedrock managed FM API Fast application building, managed model access, RAG, agents, guardrails, model choice Less low-level model infrastructure control than self-hosting
SageMaker AI hosted model Custom ML/FM lifecycle, custom training/tuning, specific deployment control More lifecycle and infrastructure decisions
Self-hosted model on compute Maximum control, special licensing, custom runtime, offline/isolated patterns Highest operations burden; pay for infrastructure whether fully utilized or not
Amazon Q / Quick / Kiro Prebuilt GenAI productivity or business applications Less custom application architecture; service-specific capabilities and pricing

Exam shortcut:

Choose managed services to move faster and reduce infrastructure work.
Choose SageMaker/self-hosting when control, customization, or lifecycle ownership is the primary requirement.

GenAI Architecture Patterns on AWS

Pattern Typical AWS services When to choose
Simple text generation app Amazon Bedrock, AWS Lambda, Amazon API Gateway, Amazon S3 Generate, summarize, rewrite, classify, or extract with minimal custom ML.
Enterprise Q&A / RAG Bedrock Knowledge Bases, Amazon Q Business, Amazon S3, OpenSearch/Aurora/Neptune/RDS vector support Need grounded answers from private or current data.
Agentic workflow Bedrock Agents, Bedrock AgentCore, Strands Agents, AWS Lambda, APIs, CloudWatch Need tool use, workflow steps, memory, and action execution.
Developer productivity Amazon Q Developer, Kiro Need code help, security scanning, modernization, specs, tests, docs.
Generative BI Amazon Quick, Amazon Q Business integration Need natural-language analytics, data stories, executive summaries, visual generation.
Custom model lifecycle SageMaker AI, SageMaker JumpStart, S3, ECR, CloudWatch Need model training/tuning/hosting control.
Secure GenAI app Bedrock Guardrails, IAM, KMS, VPC/PrivateLink, CloudTrail, CloudWatch, Macie Need privacy, access control, safety, auditability, monitoring.

Common Scenario Patterns

Scenario Best answer
"A team wants to call multiple FMs by API without managing servers." Amazon Bedrock.
"A company wants to build an internal HR assistant over documents with source citations and user permissions." Amazon Q Business or Bedrock Knowledge Bases depending whether they want a prebuilt assistant or custom app.
"A developer wants help writing code, scanning vulnerabilities, and understanding AWS services." Amazon Q Developer.
"A team wants an agentic IDE that turns prompts into specs, implementation plans, tests, and docs." Kiro.
"A BI team wants dashboards, Q&A, data stories, and executive summaries using natural language." Amazon Quick Generative BI.
"A team wants a model hub with pretrained FMs, notebooks, and deployment into SageMaker." SageMaker JumpStart.
"A team wants full ML lifecycle control for custom models." Amazon SageMaker AI.
"An agent must run in production with identity, memory, gateway tools, policy, observability, and evaluation." Amazon Bedrock AgentCore.
"Developers want an open-source SDK for autonomous agents with AWS integrations and MCP." Strands Agents.
"A customer needs safety filters, denied topics, PII masking, and grounding checks." Guardrails for Amazon Bedrock.
"The assistant is too expensive because prompts include too much history and retrieved context." Reduce input tokens, improve retrieval, summarize history, use prompt caching if supported, choose a cheaper model.
"Traffic is predictable and needs consistent throughput." Consider Provisioned Throughput.
"Traffic is low or spiky." Prefer on-demand pricing over fixed provisioned capacity.
"The model is unavailable in the required Region." Choose a supported model/Region or redesign for regional requirements.

AWS Service Anchors for Task 2.3

Category Services/features to remember
Managed foundation models Amazon Bedrock, Amazon Nova, third-party FMs in Bedrock
GenAI app building Bedrock Knowledge Bases, Agents, Guardrails, Flows, Prompt Management, Model Evaluation
Agent production platform Amazon Bedrock AgentCore Runtime, Memory, Gateway, Identity, Policy, Observability, Evaluations
Open-source agent development Strands Agents
Custom ML/FM lifecycle Amazon SageMaker AI
Pretrained model hub SageMaker JumpStart
Enterprise assistant Amazon Q Business
Developer assistant Amazon Q Developer, Kiro
Generative BI Amazon Quick
Security and compliance IAM, KMS, CloudTrail, CloudWatch, Amazon Macie, AWS Artifact, AWS Audit Manager, AWS Config
Network/data controls Amazon VPC, AWS PrivateLink, encryption at rest/in transit
Cost visibility AWS Cost Explorer, AWS Budgets, CloudWatch metrics, token metrics
Infrastructure AWS Regions, Availability Zones, Nitro, GPUs, Trainium, Inferentia

Fast Comparison Drills

Prompt Correct association
"Use FMs through an API without managing infrastructure." Amazon Bedrock
"Find pretrained FMs and deploy them in SageMaker." SageMaker JumpStart
"Train, tune, deploy, and monitor a custom ML model." Amazon SageMaker AI
"Answer employee questions from enterprise data with permissions and citations." Amazon Q Business
"Help developers generate and improve code in an IDE." Amazon Q Developer
"Prompt-to-spec-to-code workflow with agent hooks and steering files." Kiro
"Natural-language BI Q&A, data stories, and executive summaries." Amazon Quick
"Open-source SDK for autonomous agents." Strands Agents
"Managed runtime, memory, gateway, identity, policy, observability for agents." Amazon Bedrock AgentCore
"Block toxic output and mask sensitive information." Guardrails for Amazon Bedrock
"Fixed hourly capacity for predictable high throughput." Provisioned Throughput
"Cost increases because prompt contains many retrieved documents." Input token cost / context cost
"Need high availability across failures." Multi-AZ or resilient AWS architecture
"Need audit reports for compliance review." AWS Artifact
"Customer asks who secures prompts, app logic, and IAM permissions." Customer responsibility under shared responsibility

One-Page Memorization Version

  • Domain 2.3 is about AWS GenAI services, AWS infrastructure benefits, and cost tradeoffs.
  • Amazon Bedrock is the main managed AWS service for foundation models and GenAI app features.
  • Bedrock features include model access, playgrounds, Knowledge Bases, Agents, Guardrails, model evaluation, customization, and prompt management.
  • SageMaker AI is the broader ML platform for building, training, tuning, deploying, and operating models.
  • SageMaker JumpStart is the SageMaker model hub for pretrained models, FMs, examples, notebooks, and solution templates.
  • Amazon Q Business is an enterprise assistant over business data with permissions-aware responses and citations.
  • Amazon Q Developer helps developers build, understand, secure, and operate AWS applications and code.
  • Amazon Quick provides Generative BI: natural-language Q&A, data stories, executive summaries, visual and calculation authoring.
  • Kiro is an agentic coding service for specs, steering files, hooks, code, docs, and tests.
  • Strands Agents is an open-source SDK for autonomous agents.
  • Bedrock AgentCore is a managed platform for production agents: runtime, memory, gateway, identity, policy, observability, registry, evaluations.
  • AWS GenAI service advantages: accessibility, lower barrier to entry, efficiency, cost-effectiveness, speed to market, business alignment.
  • AWS infrastructure benefits: Regions, AZs, security, compliance support, shared responsibility, safety controls, observability, specialized compute.
  • Shared responsibility: AWS secures the cloud; you secure data, access, prompts, app behavior, tools, and outputs.
  • Guardrails help filter harmful content, denied topics, PII, prompt attacks, and hallucination/grounding issues.
  • Token-based pricing means input tokens and output tokens both matter.
  • Provisioned Throughput is fixed-capacity/fixed-cost and fits predictable scale or guaranteed throughput.
  • On-demand is better for variable or uncertain usage.
  • Custom models can add training, storage, hosting, evaluation, and provisioned-capacity costs.
  • Bigger models can improve quality but often increase cost and latency.
  • Region choice affects latency, data residency, compliance, model availability, and redundancy.

Mini Practice Questions

  1. A company wants to build a custom support assistant that calls foundation models by API and uses guardrails. Which service is the best anchor?

    • Answer: Amazon Bedrock.

  2. A team wants to discover pretrained foundation models, open example notebooks, fine-tune a model, and deploy it inside SageMaker. Which feature fits?

    • Answer: SageMaker JumpStart.

  3. Employees need a managed assistant that answers from enterprise data, respects document permissions, and provides citations. Which service fits?

    • Answer: Amazon Q Business.

  4. A developer wants an AI assistant to generate code, explain code, scan for vulnerabilities, and answer AWS architecture questions. Which service fits?

    • Answer: Amazon Q Developer.

  5. A product team wants an agentic coding workflow that turns requirements into specs, implementation steps, code, docs, and tests. Which service fits?

    • Answer: Kiro.

  6. A BI team wants users to ask questions of dashboards and generate data stories in natural language. Which service fits?

    • Answer: Amazon Quick.

  7. A company is building autonomous agents with an open-source SDK and wants AWS service integration and MCP support. Which tool fits?

    • Answer: Strands Agents.

  8. A company has a prototype agent and now needs secure production runtime, identity, memory, tool gateway, policy, observability, and evaluation. Which service fits?

    • Answer: Amazon Bedrock AgentCore.

  9. A chatbot needs to block harmful content, redact PII, and detect responses that are not grounded in retrieved sources. What should be added?

    • Answer: Guardrails for Amazon Bedrock, plus application validation and monitoring.

  10. A GenAI app has unpredictable low-volume traffic. Should the team start with Provisioned Throughput?

    • Answer: Usually no. On-demand is a better first fit for low or spiky traffic.

  11. A workload has predictable high-volume traffic and needs consistent throughput. What cost option might help?

    • Answer: Provisioned Throughput.

  12. A prompt includes long chat history, many retrieved chunks, and asks for a long response. Which cost drivers increase?

    • Answer: Input tokens, output tokens, latency, and quota pressure.

  13. A company needs to demonstrate AWS security and compliance reports to auditors. Which AWS service helps?

    • Answer: AWS Artifact.

  14. A question asks who is responsible for securing customer data, access permissions, and application prompts. What is the answer?

    • Answer: The customer, under the AWS shared responsibility model.

  15. A model is perfect for the task but not available in the required Region. What factor matters?

    • Answer: Regional availability, data residency, latency, and compliance requirements.

Source Anchors