[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"cheat-sheet---en":3,"domain-info---en":3,"topic-info----en":3,"prev-aws-certified-ai-practitioner-genai-fundamentals-aws-genai-stack-amazon-bedrock-en":4,"lesson-aws-certified-ai-practitioner-genai-fundamentals-aws-genai-stack-amazon-bedrock-en":19,"next-aws-certified-ai-practitioner-genai-fundamentals-aws-genai-stack-amazon-bedrock-en":536},null,{"locked":5,"reason":6,"meta":7,"item":3},true,"paywall",{"title":8,"description":9,"isFree":10,"estimatedMinutes":11,"difficulty":12,"learningObjectives":13},"Measuring Generative AI Business Value","Model scores do not prove business value. Work through the metrics the exam names, from accuracy and cross-domain performance up to ROI, conversion rate, and customer lifetime value, with a full ROI calculation on a real workload.",false,18,"intermediate",[14,15,16,17,18],"Identify the business value metrics the exam names for generative AI applications","Separate model-level metrics from operational metrics and business metrics, and explain why a high score at one level proves nothing at the level above","Define conversion rate, average revenue per user, and customer lifetime value, and name the generative AI lever that moves each one","Calculate return on investment for a generative AI application from token cost, build cost, and avoided labor","Explain why a baseline measured before launch is what makes any value claim defensible",{"locked":10,"reason":3,"meta":20,"item":31},{"title":21,"description":22,"isFree":5,"estimatedMinutes":23,"difficulty":24,"learningObjectives":25},"Amazon Bedrock","Understand Amazon Bedrock as the managed API layer for foundation models on AWS: one interface to many models, plus the built-in pieces (Knowledge Bases, Guardrails, evaluations, customization) that turn a model call into an application.",20,"beginner",[26,27,28,29,30],"Define Amazon Bedrock and explain what a fully managed foundation model service removes from your work","Explain why one API across many providers changes the cost of switching models","Identify the main built-in capabilities of Bedrock and what each one replaces","Describe how Bedrock handles customer prompts and completions with respect to model providers","Recognize which problems Bedrock is not the right answer to",{"id":32,"title":21,"body":33,"description":22,"difficulty":24,"estimatedMinutes":23,"extension":471,"infographics":472,"isFree":5,"learningObjectives":473,"meta":474,"navigation":5,"path":475,"quiz":476,"seo":533,"stem":534,"__hash__":535},"courses/courses/aws-certified-ai-practitioner/en/domains/02-genai-fundamentals/03-aws-genai-stack/01-amazon-bedrock.md",{"type":34,"value":35,"toc":464},"minimark",[36,40,43,46,51,54,57,255,262,265,268,271,275,278,285,291,359,366,372,378,384,387,391,394,397,400,403,406,410,413,416,419,422,425,429,457,460],[37,38,39],"p",{},"You have decided your application needs a foundation model. Now come the questions nobody enjoys: which GPU instances, how many, in which Region, who patches them, what happens at 3 a.m. when traffic triples, and how you swap in a better model six months from now without rebuilding everything.",[37,41,42],{},"Amazon Bedrock exists so you never answer those questions. AWS defines it as \"a fully managed service that provides secure, enterprise-grade access to high-performing foundation models from leading AI companies, enabling you to build and scale generative AI applications.\"",[37,44,45],{},"Read that as a list of things you no longer do. No servers to size. No model weights to download. No inference stack to operate. You send a request to an API endpoint and get generated text back, and the capacity underneath it is somebody else's problem.",[47,48,50],"h2",{"id":49},"one-api-many-models","One API, many models",[37,52,53],{},"Bedrock supports over 100 foundation models from providers including Amazon, Anthropic, DeepSeek, MiniMax, Moonshot AI, and OpenAI. That catalog is worth noticing, but the catalog is not the point. The point is what sits in front of it.",[37,55,56],{},"Every one of those models is reached through the same service. In practice you call the Converse API, name a model ID, and pass your messages:",[58,59,64],"pre",{"className":60,"code":61,"language":62,"meta":63,"style":63},"language-python shiki shiki-themes material-theme-lighter github-light github-dark","import boto3\n\nclient = boto3.client('bedrock-runtime', region_name='us-east-1')\nresponse = client.converse(\n    modelId='anthropic.claude-opus-4-7',\n    messages=[\n        {'role': 'user', 'content': [{'text': 'Summarize this support ticket.'}]}\n    ]\n)\n","python","",[65,66,67,80,86,139,158,176,187,244,250],"code",{"__ignoreMap":63},[68,69,72,76],"span",{"class":70,"line":71},"line",1,[68,73,75],{"class":74},"sVHd0","import",[68,77,79],{"class":78},"su5hD"," boto3\n",[68,81,83],{"class":70,"line":82},2,[68,84,85],{"emptyLinePlaceholder":5},"\n",[68,87,89,92,96,99,103,107,110,114,118,120,123,127,129,131,134,136],{"class":70,"line":88},3,[68,90,91],{"class":78},"client ",[68,93,95],{"class":94},"smGrS","=",[68,97,98],{"class":78}," boto3",[68,100,102],{"class":101},"sP7_E",".",[68,104,106],{"class":105},"slqww","client",[68,108,109],{"class":101},"(",[68,111,113],{"class":112},"sjJ54","'",[68,115,117],{"class":116},"s_sjI","bedrock-runtime",[68,119,113],{"class":112},[68,121,122],{"class":101},",",[68,124,126],{"class":125},"s99_P"," region_name",[68,128,95],{"class":94},[68,130,113],{"class":112},[68,132,133],{"class":116},"us-east-1",[68,135,113],{"class":112},[68,137,138],{"class":101},")\n",[68,140,142,145,147,150,152,155],{"class":70,"line":141},4,[68,143,144],{"class":78},"response ",[68,146,95],{"class":94},[68,148,149],{"class":78}," client",[68,151,102],{"class":101},[68,153,154],{"class":105},"converse",[68,156,157],{"class":101},"(\n",[68,159,161,164,166,168,171,173],{"class":70,"line":160},5,[68,162,163],{"class":125},"    modelId",[68,165,95],{"class":94},[68,167,113],{"class":112},[68,169,170],{"class":116},"anthropic.claude-opus-4-7",[68,172,113],{"class":112},[68,174,175],{"class":101},",\n",[68,177,179,182,184],{"class":70,"line":178},6,[68,180,181],{"class":125},"    messages",[68,183,95],{"class":94},[68,185,186],{"class":101},"[\n",[68,188,190,193,195,198,200,203,206,209,211,213,215,218,220,222,225,227,230,232,234,236,239,241],{"class":70,"line":189},7,[68,191,192],{"class":101},"        {",[68,194,113],{"class":112},[68,196,197],{"class":116},"role",[68,199,113],{"class":112},[68,201,202],{"class":101},":",[68,204,205],{"class":112}," '",[68,207,208],{"class":116},"user",[68,210,113],{"class":112},[68,212,122],{"class":101},[68,214,205],{"class":112},[68,216,217],{"class":116},"content",[68,219,113],{"class":112},[68,221,202],{"class":101},[68,223,224],{"class":101}," [{",[68,226,113],{"class":112},[68,228,229],{"class":116},"text",[68,231,113],{"class":112},[68,233,202],{"class":101},[68,235,205],{"class":112},[68,237,238],{"class":116},"Summarize this support ticket.",[68,240,113],{"class":112},[68,242,243],{"class":101},"}]}\n",[68,245,247],{"class":70,"line":246},8,[68,248,249],{"class":101},"    ]\n",[68,251,253],{"class":70,"line":252},9,[68,254,138],{"class":101},[37,256,257,258,261],{},"Now change ",[65,259,260],{},"modelId"," to an Amazon Nova model. Nothing else in that call changes. Your authentication, your SDK, your error handling, your logging, and your IAM policies all keep working.",[37,263,264],{},"Compare that to the alternative. Without Bedrock, each provider means a separate account, a separate API key stored somewhere, a separate SDK with its own request shape, a separate bill, and a separate security review. Adding a second model provider is a project. On Bedrock it is a string.",[37,266,267],{},"This is the reason the model-selection lesson could call model choice reversible. That reversibility is not a general property of generative AI. It is a property of building behind an abstraction layer, and Bedrock is that layer.",[37,269,270],{},"One honest limit: the call is portable, the behavior is not. A prompt tuned for one model does not produce identical output on another, and models differ in context window, maximum output length, and which inference parameters they accept. Bedrock makes the switch cheap. It does not make revalidation unnecessary.",[47,272,274],{"id":273},"what-bedrock-adds-around-the-model","What Bedrock adds around the model",[37,276,277],{},"A raw model call is not an application. Between \"the model responds\" and \"the feature ships\" sit a set of problems every team hits, and Bedrock ships managed answers to most of them.",[37,279,280,284],{},[281,282,283],"strong",{},"Knowledge Bases"," handle retrieval augmented generation. AWS describes the purpose directly: RAG \"uses information from data sources to improve the relevancy and accuracy of generated responses,\" and with Knowledge Bases \"you can integrate proprietary information into your generative-AI applications.\" Point it at your data, and it manages ingestion, chunking, embedding, storage, and retrieval, then returns responses with citations so an answer can be traced back to its source. A Managed Knowledge Base lets AWS run that whole pipeline; a customer-managed one lets you bring your own vector store, such as Amazon OpenSearch Serverless, Amazon Aurora, or Amazon Neptune, and control the ingestion and indexing yourself.",[37,286,287,290],{},[281,288,289],{},"Guardrails"," are the safety layer. They provide \"configurable safeguards to help you build safe generative AI applications,\" and they come as six policy types:",[292,293,294,307],"table",{},[295,296,297],"thead",{},[298,299,300,304],"tr",{},[301,302,303],"th",{},"Policy",[301,305,306],{},"What it does",[308,309,310,319,327,335,343,351],"tbody",{},[298,311,312,316],{},[313,314,315],"td",{},"Content filters",[313,317,318],{},"Detects harmful content across Hate, Insults, Sexual, Violence, Misconduct, and Prompt Attack, with configurable strength per category",[298,320,321,324],{},[313,322,323],{},"Denied topics",[313,325,326],{},"Blocks subjects you define as out of bounds for your application",[298,328,329,332],{},[313,330,331],{},"Word filters",[313,333,334],{},"Blocks exact words and phrases, including profanity and custom terms such as competitor names",[298,336,337,340],{},[313,338,339],{},"Sensitive information filters",[313,341,342],{},"Blocks or masks PII and custom regex patterns",[298,344,345,348],{},[313,346,347],{},"Contextual grounding checks",[313,349,350],{},"Flags responses not grounded in the retrieved source, or irrelevant to the question",[298,352,353,356],{},[313,354,355],{},"Automated Reasoning checks",[313,357,358],{},"Validates responses against a set of logical rules, and can suggest corrections",[37,360,361,362,365],{},"Two details about Guardrails are worth holding. They evaluate both the input prompt and the model completion, not just one side. And they can run without a model at all: the ",[65,363,364],{},"ApplyGuardrail"," API screens content on its own, so you can check text before it enters your pipeline.",[37,367,368,371],{},[281,369,370],{},"Model customization"," covers three methods. Supervised fine-tuning trains on labeled input-output examples. Reinforcement fine-tuning learns from reward functions you define instead of labeled pairs. Distillation transfers knowledge from a large teacher model to a smaller, faster, cheaper student model. Domain 3 covers when each one is the right call.",[37,373,374,377],{},[281,375,376],{},"Evaluations"," score models on your own data, either programmatically, with a judge model, or with human reviewers. You met these in the model-selection lesson.",[37,379,380,383],{},[281,381,382],{},"AgentCore"," turns a model into an agent that plans and acts. That is a topic of its own, and the next lesson picks it up.",[37,385,386],{},"The pattern across all of these is the same: each one is a problem you would otherwise solve with your own infrastructure, offered instead as a configuration.",[47,388,390],{"id":389},"the-data-question-answered-concretely","The data question, answered concretely",[37,392,393],{},"Every organization asks a version of this: if we send customer text to a third party's model, what does that third party get to see?",[37,395,396],{},"For Bedrock, AWS answers it structurally rather than with a promise. In each Region where Bedrock is available, there is one model deployment account per model provider. Those accounts are owned and operated by the Bedrock service team. When a provider delivers a model, AWS performs a deep copy of the provider's inference and training software into those accounts. And then the sentence that matters: \"Because the model providers don't have access to those accounts, they don't have access to Amazon Bedrock logs or to customer prompts and completions.\"",[37,398,399],{},"So calling a third-party model on Bedrock is not the same as calling that provider's own API. Your prompt goes to an AWS-operated account, not to the provider. The provider supplied the model; they do not receive the traffic.",[37,401,402],{},"On top of that you get the usual AWS controls: encryption with AWS KMS, private connectivity through Amazon VPC and AWS PrivateLink so requests never traverse the public internet, IAM for who may invoke which model, and CloudTrail for the audit record.",[37,404,405],{},"This is the single most useful fact in the lesson for a scenario question. When a stem mentions a regulated industry, a data residency rule, or a security review that must approve the vendor, the deployment account isolation is what the answer is built on.",[47,407,409],{"id":408},"where-bedrock-stops","Where Bedrock stops",[37,411,412],{},"Bedrock is a good default, and a default is not a universal answer. It serves and customizes models that already exist. Three situations sit outside it.",[37,414,415],{},"You want to train a model from scratch, on your own architecture, with your own training loop. Bedrock has no entry point for that.",[37,417,418],{},"You need a model that is not in the catalog, or full control over the serving environment, such as a specific inference container, a specific instance type, or deployment inside your own VPC on hardware you chose.",[37,420,421],{},"You are doing classical machine learning rather than generative AI. A fraud detection model over tabular transaction data is a gradient-boosted tree, not a foundation model, and Bedrock has nothing to offer it.",[37,423,424],{},"Each of those points at the same service, which is the subject of the next lesson.",[47,426,428],{"id":427},"exam-tips","Exam tips",[430,431,432,436,439,442,448,451,454],"ul",{},[433,434,435],"li",{},"Bedrock is the fully managed, serverless path to foundation models. Scenario words such as \"no infrastructure to manage,\" \"fastest path to production,\" or \"we do not want to run GPUs\" point here.",[433,437,438],{},"One API across many providers is the differentiator to remember. Any question about switching models cheaply, or comparing models without rewriting an application, is testing this.",[433,440,441],{},"Match the capability to the problem: proprietary data in answers means Knowledge Bases; blocking harmful or off-topic content means Guardrails; detecting an ungrounded answer means contextual grounding checks specifically.",[433,443,444,445,447],{},"Guardrails evaluate input and output, and ",[65,446,364],{}," works without invoking a model. Options claiming guardrails are output-only are wrong.",[433,449,450],{},"\"Train a model from scratch\" is the phrase that takes Bedrock off the table. So does \"we need full control of the serving container.\"",[433,452,453],{},"Provider isolation via model deployment accounts is the answer to any question about whether model providers see your prompts. They do not.",[433,455,456],{},"A customized model requires Provisioned Throughput to serve. Remember the pairing; the cost lesson explains why it stings.",[37,458,459],{},"The idea to carry forward is that Bedrock trades control for speed, and that trade is usually the right one. The next lesson is about the cases where it is not, and about the service that sits directly underneath.",[461,462,463],"style",{},"html pre.shiki code .sVHd0, html code.shiki .sVHd0{--shiki-light:#39ADB5;--shiki-light-font-style:italic;--shiki-default:#D73A49;--shiki-default-font-style:inherit;--shiki-dark:#F97583;--shiki-dark-font-style:inherit}html pre.shiki code .su5hD, html code.shiki .su5hD{--shiki-light:#90A4AE;--shiki-default:#24292E;--shiki-dark:#E1E4E8}html pre.shiki code .smGrS, html code.shiki .smGrS{--shiki-light:#39ADB5;--shiki-default:#D73A49;--shiki-dark:#F97583}html pre.shiki code .sP7_E, html code.shiki .sP7_E{--shiki-light:#39ADB5;--shiki-default:#24292E;--shiki-dark:#E1E4E8}html pre.shiki code .slqww, html code.shiki .slqww{--shiki-light:#6182B8;--shiki-default:#24292E;--shiki-dark:#E1E4E8}html pre.shiki code .sjJ54, html code.shiki .sjJ54{--shiki-light:#39ADB5;--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .s_sjI, html code.shiki .s_sjI{--shiki-light:#91B859;--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .s99_P, html code.shiki .s99_P{--shiki-light:#90A4AE;--shiki-light-font-style:italic;--shiki-default:#E36209;--shiki-default-font-style:inherit;--shiki-dark:#FFAB70;--shiki-dark-font-style:inherit}html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":63,"searchDepth":88,"depth":88,"links":465},[466,467,468,469,470],{"id":49,"depth":82,"text":50},{"id":273,"depth":82,"text":274},{"id":389,"depth":82,"text":390},{"id":408,"depth":82,"text":409},{"id":427,"depth":82,"text":428},"md",[],[26,27,28,29,30],{},"/courses/aws-certified-ai-practitioner/en/domains/02-genai-fundamentals/03-aws-genai-stack/01-amazon-bedrock",{"passingScore":477,"questions":478},70,[479,488,496,504,512,520,526],{"question":480,"type":481,"options":482,"correctAnswer":485,"explanation":487},"A team wants to call Anthropic Claude today and switch to Amazon Nova next quarter without rewriting their application. What property of Amazon Bedrock makes this possible?","single",[483,484,485,486],"Bedrock automatically translates prompts between model families","All models on Bedrock share identical training data","Bedrock exposes many models from many providers through one managed API, so switching model is a configuration change","Bedrock caches every response, so the model no longer matters","One API in front of many providers is the structural argument for Bedrock, because it turns model choice from a rebuild into a parameter. Bedrock does not rewrite your prompts for you, which is why you still revalidate after a switch, and caching is a cost feature rather than a portability one.",{"question":489,"type":481,"options":490,"correctAnswer":491,"explanation":495},"According to the AWS documentation, what access do model providers have to customer prompts and completions on Amazon Bedrock?",[491,492,493,494],"None, because providers have no access to the model deployment accounts that run their models","Read access for the purpose of improving their models","Access only in Regions where the provider operates infrastructure","Full access, which is why Bedrock requires a separate data processing agreement","AWS runs each provider's model inside a model deployment account owned and operated by the Bedrock service team, and providers have no access to those accounts, so they cannot reach Bedrock logs or customer prompts and completions. This is the concrete fact behind the claim that using a third-party model on Bedrock does not hand your data to that third party.",{"question":497,"type":481,"options":498,"correctAnswer":501,"explanation":503},"Which Amazon Bedrock capability is designed to detect responses that are not grounded in the retrieved source material?",[499,500,501,502],"Provisioned Throughput","Model distillation","Contextual grounding checks in Guardrails","Knowledge Bases reranking","Contextual grounding checks are the Guardrails filter that flags responses that are factually inconsistent with the source or irrelevant to the question, which is exactly the RAG failure mode. Reranking changes which passages get retrieved rather than judging the answer, and the other two options are capacity and cost features.",{"question":505,"type":506,"options":507,"correctAnswers":510,"explanation":511},"Which of these are Amazon Bedrock Guardrails policy types? (Choose 3.)","multiple",[323,339,508,315,509],"Token budget filters","Region restriction filters",[323,339,315],"Guardrails ships content filters, denied topics, word filters, sensitive information filters, contextual grounding checks, and Automated Reasoning checks. Token budgets and Region restrictions are real concerns on AWS, but they are handled by quotas, IAM, and service control policies rather than by a guardrail policy.",{"question":513,"type":481,"options":514,"correctAnswer":516,"explanation":519},"A developer wants to screen user-submitted text for PII before it ever reaches a model. Can Guardrails do this?",[515,516,517,518],"No, guardrails only evaluate model output","Yes, the ApplyGuardrail API evaluates content without invoking a foundation model","No, guardrails must be attached to a Bedrock model invocation","Yes, but only for models hosted on SageMaker AI","Guardrails can be attached to an inference call by ID and version, and they can also be called on their own through ApplyGuardrail, which lets you screen input or output for content that never touches a model. Guardrails evaluate both input prompts and model completions, so the claim that they only see output is wrong in both directions.",{"question":521,"type":481,"options":522,"correctAnswer":524,"explanation":525},"Amazon Bedrock is a good fit for a team that needs to train a model from scratch on their own proprietary architecture.",[523,524],"True","False","Bedrock serves and customizes existing foundation models through fine-tuning, reinforcement fine-tuning, and distillation, so it starts from someone else's pretrained model rather than from nothing. Training a novel architecture from scratch is SageMaker AI territory, which is the boundary the next lesson draws.",{"question":527,"type":481,"options":528,"correctAnswer":499,"explanation":532},"After a team fine-tunes a model in Amazon Bedrock, what must they purchase before they can run inference on it?",[529,530,499,531],"An AWS Enterprise Support plan","A dedicated Amazon Bedrock Region","A SageMaker AI endpoint","AWS states plainly that if you customized a model, you must purchase Provisioned Throughput to use it, because a customized model does not sit in the shared on-demand pool. This is the hidden cost that makes fine-tuning much more expensive than the training bill alone suggests, and the cost lesson returns to it.",{"title":21,"description":22},"courses/aws-certified-ai-practitioner/en/domains/02-genai-fundamentals/03-aws-genai-stack/01-amazon-bedrock","QWaSd16Acs3j91nQwDmm0cBJIJJUScQt-YGuxrC8e1g",{"locked":5,"reason":6,"meta":537,"item":3},{"title":538,"description":539,"isFree":10,"estimatedMinutes":11,"difficulty":12,"learningObjectives":540},"Amazon SageMaker AI and SageMaker JumpStart","Draw the line between Bedrock and SageMaker AI: what you gain when you take control of training and serving, what you take on in exchange, and where SageMaker JumpStart sits between the two.",[541,542,543,544],"Define Amazon SageMaker AI and distinguish it from the wider Amazon SageMaker platform","Compare Amazon Bedrock and SageMaker AI across control, operational burden, and the work each one removes","Explain what SageMaker JumpStart provides and why it lowers the cost of entry to SageMaker AI","Choose between Bedrock, JumpStart, and SageMaker AI for a given scenario"]