The Exam Room

Cloud Practitioner

Where the whole thing starts: what AWS sells, what it costs, and who is responsible for what.

AI Practitioner

The foundational track: what the services do and when each one fits.

AI Business Strategist

The business side of AI: cases, governance, readiness, and scaling past the pilot.

Generative AI Developer

Generative AI engineering at production scale.

Exam Room · Advanced Generative AI Developer

How to Wire an LLM to Side-Effecting Actions with Bedrock AgentCore

An assistant that has to look up a customer's subscription, pause it, refund a charge, and email confirmation. Not just answer, act. The glue between a language model and the rest of our systems comes down to one question: when the model asks to move money, what physically stops it until a human says yes? Bedrock AgentCore, a framework agent on our own infrastructure, and a hand-written tool router all put that gate in different places.

Exam Room · Advanced Generative AI Developer

Pop Quiz: OpenSearch Serverless's Hidden Floor

A Bedrock Knowledge Base needs a vector store, and on OpenSearch Serverless the idle bill depends on which collection generation you land on.

Exam Room · Advanced Generative AI Developer

Pop Quiz: When Aurora pgvector Wins

When is Aurora PostgreSQL with pgvector the better vector store for RAG?

Exam Room · Advanced Generative AI Developer

Pop Quiz: The Silent Distance-Metric Bug

Retrieval quality collapses after an embedding model swap, and nothing errors anywhere. What config is worth checking first?

Exam Room · Advanced Generative AI Developer

Pop Quiz: S3 Vectors and the Latency Budget

You have a huge, rarely-queried vector archive and a wait of a second or two is acceptable. Cheapest fit?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Why Hybrid Search Finds ERR-4021

Semantic search keeps missing exact tokens like an error code. What retrieval change fixes it?

Exam Room · Advanced Generative AI Developer

Pop Quiz: What a Reranker Actually Fixes

The right chunk is retrieved but ranks twelfth, below the cutoff. What promotes it?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Hierarchical Chunking in One Line

Answers get cut across chunk boundaries in a long structured PDF. Best Bedrock KB chunking?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Freshness and Access Are Metadata

Retrieval must respect access control and document freshness. What enforces both without a second index?

Exam Room · Advanced Generative AI Developer

Pop Quiz: The Prompt-and-Completion Record

An auditor asks to see every prompt this assistant received last Tuesday. What produces it?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Who Changed It vs What It Said

Who turned off the PII filter, and when? Which log, and why not invocation logging?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Turning Logs Into an Audit

The reviewer wants a control-mapped report and a statement of what the model is approved for. Two artefacts?

Exam Room · Advanced Generative AI Developer

Pop Quiz: The Eight Responsible-AI Dimensions

AWS names its responsible-AI dimensions. Roughly, what are they?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Packing Many Models Onto One Endpoint

How do you serve dozens or hundreds of models without an endpoint each?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Measuring Bias With fmeval

You must measure a GenAI feature for bias and toxicity before launch. Which tool?

Exam Room · Advanced Generative AI Developer

Pop Quiz: What Guardrails Enforce

What does Bedrock Guardrails actually enforce at runtime?

Exam Room · Advanced Generative AI Developer

Pop Quiz: LLM Explainability Is Traceability

Someone asks you to explain a RAG answer. What is the realistic form of FM explainability?

Exam Room · Advanced Generative AI Developer

Designing Safe Tool Schemas for an AgentCore Gateway

A gateway turns your Lambda functions and REST APIs into tools a foundation model can call directly, and every tool you publish is a capability you are granting. The arguments arrive already shaped by a model that can be steered by whatever text reached the prompt, so a tool that takes a free-form instruction lets one coaxed call reach much further than you meant. How much the schema can rule out varies with how the tool is attached, the caller's identity does not travel with the call unless you arrange it, and the confirmation gate is code you own rather than a flag you set.

Exam Room · Advanced Generative AI Developer

Pop Quiz: Faithful but Wrong

A RAG answer scores high on faithfulness but is still wrong. How?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Evaluating Both Halves of RAG

You cannot tell if a wrong RAG answer is a retrieval or a generation problem. What evaluates each half?

Exam Room · Advanced Generative AI Developer

Pop Quiz: LLM-as-a-Judge, and the Catch

You need to score thousands of outputs on quality without a human reading each. Approach and caveat?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Letting an LLM Take Actions

Answering questions is one thing; calling an internal API is another. What wires an agent for real actions on Bedrock?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Turning Down the Randomness

Outputs are too random for a structured extraction task. Which inference parameters, and which way?

Exam Room · Advanced Generative AI Developer

Finding the Documents That Never Reached the Knowledge Base

A Knowledge Base sync reports success and the assistant still cannot answer a question whose answer is sitting in one of the source buckets. The job-level view says forty thousand documents scanned and reports no failure at all, because a document that was skipped, ignored, or failed to embed leaves no trace at that altitude. Ingestion observability is a separate feature from the invocation logging most teams have already turned on, and it is the one that names the file and the reason.

Exam Room · Advanced Generative AI Developer

Building Permission-Safe Retrieval on a Bedrock Knowledge Base

For years the easy answer for a permission-sensitive corpus was Amazon Kendra, which crawled document ACLs alongside document content. Kendra closed to new customers on 30 July 2026. A Bedrock managed knowledge base now crawls those permissions too, on the connectors that have one, and filters retrieval against an identity the application supplies; where a source falls outside that list, enforcement goes back to metadata filters the team designs. This is what each path involves, and where the line between them falls.

Exam Room · Advanced Generative AI Developer

Pop Quiz: Provisioned Throughput vs On-Demand

Steady high-volume Bedrock traffic with a latency commitment. Reserve capacity or stay on-demand?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Prompts Are Versioned Assets

Thirty services share prompts and you need versioning and reuse. What on Bedrock?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Cutting the Bill Without Losing Quality

Your Bedrock bill is high but quality must hold. First levers?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Seeing Into a Production Bedrock App

What gives you operational visibility into a production Bedrock app?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Keeping PII Out of Prompts and Logs

Keep customer PII out of prompts and logs. What is the built-in control?

Exam Room · Advanced Generative AI Developer

How to Pay for Serving a Model on Bedrock

Three models are heading for production: a hosted foundation model, a fine-tune halfway through training, and a set of open weights from a research spike. Finance wants one forecast. Each bills on a different unit, and for two of them the unit was settled the moment somebody picked a base model. Serving cost is less a lever you pull afterwards than a constraint you inherit from where the weights came from.

Exam Room · Advanced Generative AI Developer

Choosing an Agent Framework for the AgentCore Runtime

Deciding to own the reasoning loop settles where the agent runs and leaves open what you write it in. AgentCore takes any framework, so the choice is not about compatibility; it is about how much of the runtime each one already implements, whether the traces arrive without you building them, and what happens when one agent becomes four. Frameworks that already emit OpenTelemetry with the GenAI conventions make observability a dependency rather than a project, and that separates the field more than the loop syntax does.

Exam Room · Advanced Generative AI Developer

Giving an Agent Credentials Without a Standing Key

An agent that reaches real systems needs credentials, and the quickest way to give it some is the one that ages worst: a long-lived key in the environment that every request shares. The alternative splits into two questions that get conflated. Establishing who is asking is inbound authorisation and happens before your code runs; obtaining something to call a downstream service with is outbound, and which flow you can use depends on whether the call is machine-to-machine, delegated by a user, or exchanged from a token you were already handed.

Exam Room · Advanced Generative AI Developer

Picking an Evaluation Metric From the Cost of Being Wrong

A support assistant reports 97% accuracy every week while one of its two classifiers keeps approving fraudulent refunds. Both flag rare events, both look healthy on a single quality number, and the expensive error runs in opposite directions: silencing a real customer on one, paying out fraud on the other. Choosing a metric starts by naming the positive class and the error that costs, then reading the answer off the four cells of a confusion matrix rather than off the wording of the problem.

Exam Room · Advanced Generative AI Developer

Reviewing a GenAI Workload Against the Generative AI Lens

Two weeks before a subscriber-facing assistant goes live, the platform group asks for a written architecture review. A general one will find real problems and still never ask whether the model was chosen against an evaluation set, whether retrieved text can carry instructions, or what happens when the model version is retired. Those questions come from a lens, and the lens has somewhere to put the answers.

Exam Room · Advanced Generative AI Developer

Switching Foundation Models Without Shipping Code

Eight services call Amazon Bedrock and every one of them has the model id compiled into it. A cheaper model lands, and moving to it takes eight pull requests, eight reviews and eight release trains. The values that change fastest are trapped in the artefact that changes slowest, and getting them out is a design decision with more options than it first looks.

Exam Room · Advanced Generative AI Developer

Promoting a Fine-Tuned Model into Production

The training job finished and the eval numbers beat the base model. That was the easy half. The artefact now has to serve real traffic, stay identifiable two years later, and come back out inside ten minutes when it misbehaves, and the four places a customised model can live answer those three demands very differently.

Exam Room · Advanced Generative AI Developer

One Vector Index or Many

Forty million chunks, six business domains, eleven tenants, one OpenSearch index, and a p99 that has drifted past the budget. Splitting the index is the obvious move, and it is a memory decision, a rebuild-blast-radius decision and a tenancy decision at the same time. Index topology decides more here than another round of graph-parameter tuning.

Exam Room · Advanced Generative AI Developer

Building a Deployment Pipeline for a GenAI Feature

A generative feature is not one deployable binary. It is a prompt version, a guardrail version, an agent alias, a knowledge base sync and an inference profile ARN, and they have to promote together or a rollback returns half a release. Building the pipeline means deciding what a release is, then finding a gate that can fail on a score instead of an assertion.

Exam Room · Advanced Generative AI Developer

Putting a GenAI Gateway in Front of Bedrock

Eleven services call Amazon Bedrock directly, and the platform team can answer none of the questions it keeps being asked: who spent what, who is about to exhaust the account quota, and which callers skipped the guardrail. Routing every request through platform-owned compute answers all three. It also adds a hop of latency, a new tier-one dependency, and the loss of the caller identity that used to arrive at Bedrock intact.

Exam Room · Advanced Generative AI Developer

Choosing Where an MCP Server Runs

Three tools to publish to one agent, all speaking the same protocol: a subscriber lookup that reads a single row, an analytics query that answers out of a 400MB index it holds in memory, and a route optimiser that shells out to a licensed binary. MCP settles how the agent calls them and says nothing about where any of them runs. What each server keeps between calls decides that.

Exam Room · Advanced Generative AI Developer

Putting Brakes on an Autonomous Agent

An invoice-reconciliation agent looped forty times against a supplier API that was returning 500s, spent a night's token budget getting nowhere, and did it under a tool role that could read every object in the bucket. None of that is a model failure. It is a loop whose length nobody decided, running with permissions nobody scoped, and the fix is a set of bounds enforced outside the model, by code that counts.

Exam Room · Advanced Generative AI Developer

Wiring a GenAI Assistant Into Systems You Cannot Change

The order system is fifteen years old, speaks SOAP, falls over above a handful of requests a second, and goes dark for three hours every night. Nobody is allowed to change it. The assistant still has to answer questions about what is in it, which makes the design question how the data gets across the gap rather than what the model does with it once it arrives.

Exam Room · Advanced Generative AI Developer

Serving a GenAI Feature When the Data Cannot Leave

A regulator says the plant's maintenance records never leave the site, and a technician on the factory floor wants an answer before their thumb comes off the button. Neither requirement moves. The design job is to work out which bytes are actually pinned, which milliseconds are actually felt, and what is left over that can safely be sent to a foundation model in a Region.

Exam Room · Advanced Generative AI Developer

Detecting Misuse of a Public GenAI Assistant

The assistant has logging. What it does not have is detection, or any answer to what happens automatically once a pattern turns up. Three identities are already misbehaving in three different ways, and each one shows up in a different control. Choosing where the signal comes from is half the work; choosing how far the automated response may go is the half that gets skipped.

Exam Room · Advanced Generative AI Developer

Deleting a Subscriber's Data From a RAG System

One subscriber asks to be deleted, and the record turns out to sit in nine places at once: source documents, a vector index, conversation state, two sets of logs, a response cache, an evaluation set, a tuning dataset, and the weights of the model currently answering traffic. Three strategies cover that ground, and which one you can use was decided long before the request arrived.

Exam Room · Advanced Generative AI Developer

Keeping a Vector Store Healthy in Production

Retrieval p99 has tripled in three months, the answers have got worse, and the store has not logged a single error the whole time. Three different failures produce that same symptom: the collection is saturated, the index no longer fits the corpus it was built for, or the data in it is wrong. They need three different remedies, and the wrong one uses up a maintenance window and leaves the latency where it was.

Exam Room · Advanced Generative AI Developer

Auto-Scaling a Model Endpoint for Bursty GenAI Traffic

An internal assistant runs an open-weight model on a SageMaker real-time endpoint, and every weekday brings the same shape: a wall of traffic between 08:15 and 09:30, then near-silence until the next morning. Scale-out lands ten minutes after the burst starts, so the fleet has been pinned at peak size around the clock to hide the lag. Getting that money back means scaling on a signal that tracks what the accelerators are actually holding, and giving the burst somewhere to wait while capacity arrives.

Exam Room · Advanced Generative AI Developer

Dashboards for a GenAI Feature: Operations, Quality, and Business

Three people ask for a dashboard for the same generative-AI assistant in the same week, and they mean three different things. The on-call engineer wants a number that moves in seconds and can raise an alarm. The product owner wants yesterday's figures joined to the bill. The compliance reviewer wants a record they can query, and a chart will not do. One screen serving all three is how a team ends up with a screen nobody watches.

Exam Room · Advanced Generative AI Developer

Evaluating an Agent's Run, Not Just Its Answer

A refund agent gets the number right about four times in five, and nobody can say what happened in the other one. Answer-correctness scores a string. An agent produces a trajectory: a sequence of tool choices, arguments, results and retries, any step of which can be the thing that broke. Scoring the run instead of the reply is what turns "it fails sometimes" into a defect you can name.

Exam Room · Advanced Generative AI Developer

Turning a Golden-Set Score Into a Deployment Gate

A team has 400 golden examples, a working Bedrock evaluation job, and a habit of merging prompt changes because whoever wrote them read a dozen answers and liked what they saw. The last two regressions were found by customers. Turning that score into something that can stop a release means deciding what runs where, what a failure costs, and what happens when the number lands close to the line.

Exam Room · Advanced Generative AI Developer

Catching a Regression After the Deploy, Not From the Complaints

Latency was flat, the error rate was zero, and the bill had not moved. The answers had still got worse, and nobody noticed for three days. Deployment validation for a generative feature has to assert on the content of an answer, on a schedule, against production, because the changes that ruin quality often arrive without a deploy at all.

Exam Room · Advanced Generative AI Developer

Which Bedrock Errors to Retry and Which to Surface

A Bedrock feature fails about two percent of its calls, the application log says 'model call failed' and nothing else, and every invocation is wrapped in a retry-three-times loop. Amazon Bedrock names every failure, and the name separates the ones worth retrying from the ones that will fail identically forever; a wrapper that treats them the same turns a permanent error into a slow permanent error and turns a quota ceiling into an outage the team caused itself.

Exam Room · Advanced Generative AI Developer

Finding Out Why a Prompt Stopped Behaving

A summariser that reliably returned JSON starts returning prose about one call in six, nobody has touched the template, and the console reproduces the good behaviour every single time. The template is only part of the prompt. The rest is variables, retrieved passages and conversation history, assembled at request time and written down nowhere. Until those bytes are recorded there is nothing to debug.

Exam Room · Advanced Generative AI Developer

Getting Evaluation Results in Front of the People Who Decide

Three model comparisons in a quarter, each one answered in a chat thread with a pasted table, and the person paying for the work still cannot say which model the feature runs on. Results that live in the team's scrollback have been produced but never delivered. Where they should land depends on who is reading them and what that person is about to decide.

Exam Room · Advanced Generative AI Developer

Cheat Sheet: Integration and Deployment

A revision sheet for the half of Domain 2 that is not agents: where a model actually runs, how the answer reaches the caller, how a generative feature bolts onto systems nobody will rewrite for it, and what keeps the whole thing standing once the quota runs out.

Exam Room · Advanced Generative AI Developer

Pop Quiz: Backoff or Breaker

Nine calls in ten are failing and every request waits out the whole retry ladder before giving up. More backoff, or a breaker?

Exam Room · Advanced Generative AI Developer

Pop Quiz: What Model Registry Versions

Bedrock versions prompts, guardrails and agents. None of those describes a training artefact. What does a model package version actually hold?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Expand, Decompose, or Transform

Short queries in customer language miss; the same question in the documentation's words lands. Hybrid search and a reranker have already been tried. What now?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Which Amazon Assistant

Three asks land in the same week: engineers refactoring a legacy Java service, staff searching SharePoint and the ticketing system, and a customer-facing assistant inside the company's own product. Which surfaces?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Strands, Agent Squad, or AgentCore

One agent becomes four, and something now has to send each message to the right one. Which piece do you add, and which stay as they are?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Where the MCP Server Lives

Two MCP servers, one agent. One wraps a DynamoDB lookup and answers in 40ms. The other spends ninety seconds building a 6GB in-memory graph, then answers path queries against it. Where does each run?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Getting a GenAI Feature in Front of Users

Three asks in one week: a hosted chat UI with auth, an API contract agreed before anyone codes, and a pipeline reorderable without a deployment. Three different tools.

Exam Room · Advanced Generative AI Developer

Pop Quiz: The Rule That Has to Be Provably Followed

An insurance assistant has to prove its eligibility answers follow from the published policy. Which guardrail policy carries that, and why is grounding not enough?

Exam Room · Advanced Generative AI Developer

Pop Quiz: The Bias Number That Went Stale

A hiring-summary feature passed its fairness evaluation nine months ago. Four prompt edits and two model versions later, nobody has re-measured. What keeps the number current?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Proving the Knowledge-Base Bucket Is Not Shared

Sign-off needs proof that nobody outside the account can read the knowledge-base bucket or use the key over the vector index. Which control produces it, and keeps producing it?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Naming the LLM Risk in a Pen-Test Finding

Four findings against one assistant, four different layers to fix them in. Which control closes the tool that can cancel anyone's subscription?

Exam Room · Advanced Generative AI Developer

Pop Quiz: A Bedrock Bill That Doubled Overnight

A seasonal Bedrock feature grows every month, so a fixed budget threshold never fires or fires constantly. What tells you the daily curve has left its normal?

Exam Room · Advanced Generative AI Developer

Pop Quiz: What to Scale a Model Endpoint On

CPU utilisation sits at nine per cent while the morning burst queues for minutes. What signal should the endpoint scale on, and why is the reflex metric wrong for token generation?

Exam Room · Advanced Generative AI Developer

Pop Quiz: The Agent Called the Wrong Tool, or the Tool Failed

The agent's answers got worse this week and its own error rate never moved. Did the model start selecting a different tool, or did a tool start returning empty results?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Retrieval Got Slow and the Answers Got Worse

Latency has tripled, compute is under its ceiling, ingestion is green, and recall has slid from 96 to 88 per cent. Three things go wrong with a vector store. Only one fits.

Exam Room · Advanced Generative AI Developer

Pop Quiz: Right Answer, Wrong Route

A refunds agent scores 92% on final-answer correctness and operations keep finding refunds issued twice. What do we measure next?

Exam Room · Advanced Generative AI Developer

Pop Quiz: Reading a Bedrock Exception

A nightly batch now takes six hours and still fails. The logs hold two exception names, and only one of them is worth retrying.

Exam Room · Advanced Generative AI Developer

Pop Quiz: The Gate That Blocks on Noise

Every pull request runs 400 golden examples through an evaluation job and hard-fails below 0.8. Builds take fifty minutes and three of the four blocks passed on a re-run. What changes?

Exam Room · Advanced Generative AI Developer

Flash Card: AWS Cost Anomaly Detection

Machine-learning detection of unusual spend against a learned baseline, free to run, and a feature within AWS Cost Explorer.

Exam Room · Advanced Generative AI Developer

Flash Card: Amazon Managed Grafana

A managed Grafana workspace for querying and visualising metrics, logs and traces from several sources on one dashboard, with sign-in through IAM Identity Center and pricing per active user.

Exam Room · Advanced Generative AI Developer

Flash Card: Amazon CloudWatch Synthetics

Scripted canaries that call a live endpoint on a schedule and publish success rate and latency as CloudWatch metrics, so a quiet GenAI feature still produces a signal when nobody is using it.

Exam Room · Advanced Generative AI Developer

Flash Card: Amazon Bedrock Model Evaluations

Amazon Bedrock evaluations: automatic metrics, a judge model, a human work team you bring, and the separate retrieval and response scores a RAG evaluation job reports.

Exam Room · AI Practitioner

Telling AI, ML, Deep Learning, and Agentic AI Apart

Five systems, one governance review, five labels to assign. A rules engine, a churn model, an image checker, a summariser and a tool-calling agent are not the same kind of thing, and the label sets who reviews each one, what data has to be produced, and what kind of explanation the business can give.

Exam Room · AI Practitioner

Cheat Sheet: ML Fundamentals and the SageMaker Suite

A condensed revision sheet for the classic machine-learning ground the AI Practitioner exam assumes: the learning-problem taxonomy, overfitting and drift, and the SageMaker services that get confused for one another.

Exam Room · AI Practitioner

Turning a Business Question Into an ML Problem

Six departments in a produce-delivery business arrive in the same fortnight with six asks, and every one of them gets introduced with the words 'can AI do this'. The route from a business question to a technique runs through two smaller questions: what does the answer look like, and has anyone written down the right answer before. Settle those and the AWS service follows from the answer.

Exam Room · AI Practitioner

Mapping an AI/ML Pipeline Onto AWS Services

A churn model that works in a notebook and a Bedrock summariser that works in a demo both have to become something a business can run. Drawing that on one page means naming the stages of an AI/ML pipeline without collapsing four data stages into one, and putting an AWS service against each. The two pipelines share a shape and differ in the middle.

Exam Room · AI Practitioner

Traditional Model or Foundation Model

A lender has two features on the same backlog: a credit decision that has to be explained to the applicant it refused, and an email triage feature with no labelled history and a fortnight to ship. One of them needs a model trained on the lender's own data. The other needs a model somebody else already trained. The split falls out of what has to be explained and what has already been labelled.

Exam Room · AI Practitioner

What Your Data Decides Before You Pick a Model

Four datasets land on one team in the same week: a subscriptions table in Amazon Redshift, three years of daily delivery counts, a folder of scanned delivery notes, and 400,000 support emails nobody has ever sorted. Only two of them can become a model this quarter, and the algorithm has nothing to do with it. What the data already is, and how many of its examples carry an answer, settles which learning method is available at all.

Exam Room · AI Practitioner

When a Prediction Is the Wrong Answer

Four proposals land in the same budget round at a box-delivery business, and all four arrive described as AI projects. Two of them need a model. The other two need a refund calculation and a statutory eligibility check to come out the same way every time, which is a rule, not a prediction, however good the accuracy figure on the slide looks.

Exam Room · AI Practitioner

How Much MLOps a Model Actually Needs

A churn model has been scoring subscribers every night for eight months, is getting worse with nothing measuring it, and cannot be rebuilt because nobody knows which folder of training data produced it. How much operational machinery to put around a model is a real decision with three sensible answers, and the smallest one is right more often than engineers like to admit.

Exam Room · AI Practitioner

Measuring Whether a Model Earned Its Keep

A recommendation model arrives at the quarterly review with an F1 score of 0.81, up from 0.74, and the finance lead asks the one thing nobody prepared for: is it worth what we are paying for it. Model performance metrics say whether the model is right. Business metrics say whether being right returned more than it cost, and the two can move in opposite directions.

Exam Room · AI Practitioner

Choosing How a Model Serves Its Predictions

One trained fraud model, four teams asking to call it, and four completely different jobs behind the same request. A checkout page needs a score in 200ms, an overnight job needs 2.4 million scores with nobody waiting, an analyst screen runs sixty times a day, and a document pipeline needs several minutes on a 400MB upload. Real-time, serverless, asynchronous and batch are four shapes of job rather than four sizes of the same one.

Exam Room · AI Practitioner

Which Stage of the Foundation Model Lifecycle Is Yours

A retailer with 9,000 product pages and no descriptions has three proposals on the table: prompt a model on Amazon Bedrock, fine-tune one on four years of house copy, or train one on the catalogue from nothing. They are three entry points into the same seven-stage sequence, and what separates them is which stages the team signs up to own.

Exam Room · AI Practitioner

What Belongs in the Context Window

A support assistant on Amazon Bedrock gets answers wrong in two ways: sometimes it does not know the refund policy, and sometimes it forgets what the customer said four turns ago. Both are the same shortage. Everything a model sees on one call arrives through one finite budget of tokens, and deciding what fills that budget is context engineering.

Exam Room · AI Practitioner

When an AI Application Becomes Agentic

An internal travel assistant answers policy questions well and fails at every booking, because one call to a model can only write about acting. Giving it tools turns that call into a loop whose next step is not fixed in advance, and the loop changes what the system costs, how predictable it is, and whether anyone can reconstruct afterwards what it did.

Exam Room · AI Practitioner

What a Token Costs and What Changes the Bill

A claims team put a Summarise button on twelve-page documents and got a bill four times the estimate. Tokens are what a foundation model reads and writes, and tokens are what Amazon Bedrock meters, so the length of the prompt, the length of the answer, and the number of times the same document gets sent again are the three things moving the number.

Exam Room · AI Practitioner

Where Generative AI Helps and Where It Hurts

An insurer has three features on one backlog: a claims-summary drafter, a policy-eligibility decision, and a customer chat assistant. The properties that make a generative model adaptable and conversational are the same ones that make it nondeterministic and hard to interpret, and they do not come apart. What separates the three features is whether a wrong answer is embarrassing or actionable.

Exam Room · AI Practitioner

The AWS Building Blocks for a GenAI Application

Four teams ask for generative AI in the same week and each of them means something different by it. Amazon Bedrock, SageMaker AI, JumpStart, Amazon Quick, Kiro, Strands Agents and Bedrock AgentCore are seven building blocks, and two questions sort them: who the output is for, and how much of the stack you want to run yourself.

Exam Room · AI Practitioner

Proving a GenAI Feature Paid for Itself

Six weeks after a retailer shipped a shopping assistant on Amazon Bedrock, the finance review wants to know whether to keep paying for it, and the only numbers anyone has prepared are model-quality scores. Those say the model works. Whether the business is better off is a different set of numbers, measured against a baseline somebody had to capture before launch.

Exam Room · AI Practitioner

What to Weigh When You Pick a Foundation Model

A clinic-software company is adding three generative features at once, and the instinct is to wire the strongest model in the catalogue into all three. Model type, performance requirements, capabilities, constraints, compliance, cost, latency and model complexity are eight separate filters, and they rarely agree on one answer. Run them per job and the shortlist stops being a leaderboard.

Exam Room · AI Practitioner

Answering Questions From Your Own Documents

A regional insurer wants its staff to ask plain questions of four thousand policy documents and get back an answer with a source they can open. A foundation model on its own cannot do that: the documents were never in its training data, and they do not fit in a prompt. Retrieval Augmented Generation closes the gap, and the decision that follows is where the vectors live.

Exam Room · AI Practitioner

Deciding How Far to Customise a Foundation Model

A veterinary practice assistant reads clinical abbreviations wrong, quotes last year's medicine schedule, and answers in a tone the practices dislike. Three teams have proposed three different fixes, and a fourth voice wants to train a veterinary model from scratch. Six ways to customise a foundation model sit on a cost ladder, and which failure you actually have decides how far up it you need to climb.

Exam Room · AI Practitioner

Picking a Prompting Technique for the Task

Four prompts in one logistics product fail in four different ways: one rambles, one invents categories that do not exist, one gets multi-step arithmetic wrong, and one keeps drawing text into the picture. Adding more examples fixes exactly one of them. Choosing a prompting technique starts with naming which part of the prompt is missing.

Exam Room · AI Practitioner

Choosing a Foundation Model for the Job

A travel booking company ships three model-backed features in one release: a chat assistant answering in three languages, an overnight job that reads two-hundred-page contracts, and alt text for four hundred thousand hotel photographs. Modality, latency, multi-lingual coverage, input/output length and cost are separate filters, and they disagree with each other. Run them per feature, then set the inference parameters on purpose.

Exam Room · AI Practitioner

Cheat Sheet: Generative AI Foundations

A condensed revision sheet for the generative-AI groundwork that comes before any Bedrock service enters the picture: the architectures behind the models, the vocabulary around them, their known limitations, and the prompting techniques that sit underneath all of it.

Exam Room · AI Practitioner

Judging Whether a Foundation Model Is Good Enough

A news site wants to move its article summaries onto a cheaper model, and nobody can say whether quality would drop. Three different questions are hiding inside "is it good enough": does the output match a reference answer, would a person call it good, and does it clear the bar the job needs. Each one has its own measurement, and reaching for the wrong one is how a summariser gets scored with a translation metric.

Exam Room · AI Practitioner

Which Kind of Training a Foundation Model Needs

A marine insurer has exhausted the cheap fixes and accepted that the model itself has to change. Four training jobs are on the table, and they eat completely different data: raw text, labelled pairs, a bigger model's answers, or human rankings. Which job to run is settled by which data the company actually holds and can afford to prepare.

Exam Room · AI Practitioner

When an AI Agent Earns Its Place

An equipment-hire company wants three jobs automated, and one proposal has been made for all three: build an agent. Only one of the three has steps that nobody can write down before the request arrives, and that is the one an agent suits. The other two come out cheaper, faster and far easier to explain without one.

Exam Room · AI Practitioner

Cheat Sheet: AWS AI Services

A condensed revision sheet for the AWS AI service list: one line each for the purpose-built services, the generative build surface and the named Bedrock features, plus the pairs that get mixed up under pressure.

Exam Room · AI Practitioner

Keeping Prompts Safe and Versioned

Three incidents in one month at a retail bank: the assistant printed its own system prompt into a customer chat, a pasted support macro talked it out of its refusal rules, and a Friday wording change had it quoting fee amounts nobody could source. The prompt was a string literal in a Lambda, so there was nothing to roll back to. Naming the four risks correctly is where the repair starts.

Exam Room · AI Practitioner

Measuring Whether an AI Feature Is Working

Three months after a telco shipped its support assistant, the model scores are healthy, deflection is up, and the head of support wants the thing switched off. Scores that grade an answer against a reference say nothing about whether the feature is worth running. Those are two different measurements, and the budget review turns on the second.

Exam Room · AI Practitioner

Deciding Whether a Dataset Is Fit to Train On

Forty thousand support tickets, four years of them, filed into twelve queues by three people who never once compared notes. Training a triage model on that export takes a fortnight of compute and contractor time, and nobody has checked whether the data can carry a model at all. Two unrelated defects hide in a set like this, and each needs a different fix.

Exam Room · AI Practitioner

Explaining an AI Decision to the Person It Affects

A storm-damage claim is declined and three people ask why: the claimant, the compliance lead, and the developer who spotted a flood clause in a letter about a patio roof. Same word, three different artefacts, and handing any of them the wrong one produces something nobody can act on.

Exam Room · AI Practitioner

When the Decision Has to Be Explainable

A lender has to decline some applications and tell each applicant why. That obligation reaches back into the model choice: the more accurate model produces no reason a letter can use, and the model whose reasons are legible is four points worse. AWS asks you to name that trade, and it is settled with numbers rather than preference.

Exam Room · AI Practitioner

Reducing the Legal Exposure of a Generative Feature

A retailer wants generated product descriptions and campaign images across 60,000 lines, and legal has refused to sign it off. Counsel's memo lists five worries, and they close in three different places: a contract, a pipeline, and a person. Sorting which is which is most of the work.

Exam Room · AI Practitioner

Keeping Watch on Bias After Launch

A loan-triage model passed its fairness check six months ago and nobody has looked since. Overall accuracy has not moved all year, which is why nobody noticed that one cohort's approval rate fell sixteen points. Building the ongoing watch means deciding which instrument covers which part of the job, and how often each one has to run.

Exam Room · AI Practitioner

Diagnosing a Model That Works for Most People

A delivery-window model hits 94 per cent overall and misses badly for forty outer postcodes and for everyone over seventy. Three things cause that, and their repairs pull in opposite directions, so guessing deepens the failure. The diagnosis comes out of one comparison run twice: training against validation error, overall and then slice by slice.

Exam Room · AI Practitioner

Picking a Model When Sustainability Is on the Scorecard

A building-services contractor has a published emissions target and a new summarisation feature to build, and procurement wants the energy cost of the model choice written on the same page as the accuracy and the bill. Model size, tokens processed and idle capacity are the three things that move that number, and all three are settled before a single request is served.

Exam Room · AI Practitioner

Matching an AI Security Worry to an AWS Control

A small insurer's security review of its Bedrock assistant produces six worries and no agreement on what fixes what. An identity problem, a network path, an uncatalogued bucket, an encryption question, a model that talks about competitors, and an agent about to call the refunds API. Each has one control that answers it, and several that sound like they might.

Exam Room · AI Practitioner

Sorting AI Security Risks Into the Layer That Owns Them

Eight incident reports from one quarter of running a retail assistant, and every review meeting stalls in the same place: somebody proposes a fix before anyone has said what kind of problem it is. A four-year-old library with a CVE, an insult in a reply, an order number belonging to a different customer. Sorting each one into the layer that owns it decides the fix.

Exam Room · AI Practitioner

Cheat Sheet: Responsible AI, Security, and Governance

A condensed revision sheet for the two smallest-weighted but most-confused domains: the eight AWS responsible-AI dimensions with their canonical scope, and the security and governance vocabulary that sits underneath every AI workload, not just a generative one.

Exam Room · AI Practitioner

Choosing the Service That Produces the Evidence

An assessor asks a healthcare startup five questions about its Bedrock summarisation feature and gives it a fortnight. Five different AWS services answer them, and the team keeps reaching for the wrong one because all five sound like 'the compliance service'. What separates them is the object each one inspects and who wrote the evidence down.

Exam Room · AI Practitioner

Keeping an Assistant From Making Things Up

A council assistant has invented a bin-collection rule, quoted a fee that stopped applying two years ago, and cited a real policy document for something that document does not say. Three complaints in a fortnight, three different faults, and a fix for one of them that does nothing at all for the other two.

Exam Room · AI Practitioner

Four Things a Dataset Has to Be Before a Model Sees It

Eight years of job applications, and four people in one meeting each want something different to be true of them before anyone fine-tunes anything. Is the data right, is it safe to hold, who may read it, and is it still what it was are four questions, not one. Sorting a requirement into the right one decides which AWS service answers it.

Exam Room · AI Practitioner

Deciding Where AI Data Lives and How Long It Stays

Legal sends three questions about a Bedrock study assistant and the team cannot answer any of them: where does student data physically sit, how long do the conversation logs stay, and how would anyone notice if either answer stopped being true. A generative AI feature leaves a fresh copy of the data at every stage, and each copy has its own home, its own clock, and its own way of outliving the policy.

Exam Room · AI Practitioner

Setting Up Governance Before the First AI Feature

Four AI uses land on a 200-person logistics company in the same month, and one operations manager is told to have governance in place before any of them ships. She does not need four programmes. She needs a register, a one-page policy, and a rule that decides which of the four has to pass which review, and how often it comes back.

Exam Room · AI Practitioner

Showing Where an AI System's Data Came From

A regulator disputes one sentence a research assistant produced, and four people in the same company ask where did this come from in four differently shaped ways. One wants the passage, one wants the pipeline run, one wants the list of datasets and their owners, and one wants the model's build history. Four questions, four artefacts, and reaching for the wrong one wastes a fortnight.

Exam Room · AI Practitioner

Pop Quiz: A Rule, Not a Prediction

A classifier that decides which allergen warnings go on the label, 99.4% accurate on four years of past labels. The accuracy figure is doing all the persuading, and it is the wrong number to be persuaded by.

Exam Room · AI Practitioner

Pop Quiz: Batch or Asynchronous

300MB scanned bundles, eight minutes each, a few dozen a day at no predictable hour, and the uploader wants a notification when it is done. Two of the four inference options are described as 'not real-time', and only one of them fits.

Exam Room · AI Practitioner

Pop Quiz: No Labels, No Target

Three years of subscriber behaviour, no labels, no target column, and a marketing team who want to know whether there are natural groups worth treating differently. Four of the five techniques on offer need something the data does not have.

Exam Room · AI Practitioner

Pop Quiz: Which Managed AI Service

Recorded calls to text, transcripts checked for personal data and scored for sentiment, summaries in Spanish, and a voice bot on the front door. Four jobs, four services, and two pairs that get swapped more often than any others.

Exam Room · AI Practitioner

Pop Quiz: Managed API or Self-Hosted

A proprietary model the provider offers only through an API, four application developers and nobody who runs ML infrastructure, and a few hundred summaries a day arriving in two bursts. Three of the five routes would leave the insurer running GPU instances nobody asked for.

Exam Room · AI Practitioner

Pop Quiz: The Model Is Not Wrong, the World Moved

A demand model accurate at launch has been getting steadily worse for fourteen months. Same columns, same input distributions, and nobody has touched the pipeline. What did change was the business.

Exam Room · AI Practitioner

Pop Quiz: Ninety-Nine Percent and Blind

A fraud classifier reports 99.2% accuracy. The fraud team says nothing is being caught. Both are true at once, and the reason is printed in the dataset rather than in the model.

Exam Room · AI Practitioner

Pop Quiz: Generative or Agentic

Three systems in one support team, all on the same kind of model, and only one is agentic AI. Sorting them takes one test: does the system act, or hand its output to somebody who acts?

Exam Room · AI Practitioner

Pop Quiz: Which Job Suits a Foundation Model

Four projects land on one desk the same week: catalogue images and a promo clip, call summaries, email routing into eleven queues, and a shopping assistant. Three are foundation-model work. One only looks like it.

Exam Room · AI Practitioner

Pop Quiz: Input Tokens, Output Tokens, and the Bill

A summarisation service sends six thousand tokens of document and gets two hundred tokens of summary back, and the bill tracks the documents rather than the summaries. The team expected the opposite. One sum explains it.

Exam Room · AI Practitioner

Pop Quiz: What MCP Is For

An agent that reaches a ticketing system, a document store and an HR API has three hand-written integrations behind it, and a fourth system means a fourth. Model Context Protocol stops the count growing that way, and it is not the four things it gets mistaken for.

Exam Room · AI Practitioner

Pop Quiz: Naming the Generative AI Limitation

An audit team runs one prompt through the same model three times and gets three different answers. All three are correct. Naming the limitation is harder than it looks: the word most people reach for is the one that does not apply.

Exam Room · AI Practitioner

Pop Quiz: Which Metric Convinces Finance

An ecommerce team has a ROUGE score, a human-preference win rate, and six weeks of a live recommendation assistant. The CFO wants to know whether to fund it again. Both of those numbers describe the model; neither describes the return.

Exam Room · AI Practitioner

Pop Quiz: Bedrock, SageMaker AI, or JumpStart

A named open-weight model, a VPC it has to run inside, instance types the team wants to choose, and no training at all. Five options from the same shortlist, and only one of them puts those weights on infrastructure you pick.

Exam Room · Advanced Generative AI Developer

Cheat Sheet: Evaluation Metrics

A condensed revision sheet for every evaluation metric across a generative AI system: classification, ranking, regression, and text metrics, sorted by the output shape they score and the question that picks one over another.

Exam Room · AI Practitioner

Pop Quiz: Which Stage of the FM Lifecycle

The assistant shipped a month ago and users are clicking thumbs-up and thumbs-down under every answer. Somebody in the stand-up asks what that activity is called and where it goes. Name the stage.

Exam Room · AI Practitioner

Pop Quiz: Context Engineering or Fine-Tuning

An internal assistant answers refund questions with a rule that changed last month. One team wants to fine-tune, another wants a bigger model. Neither is cheapest.

Exam Room · AI Practitioner

Pop Quiz: Facts That Change Weekly

Delivery windows and stock policy get rewritten every Monday, and the assistant quotes last month's. The team has costed a monthly fine-tuning run. A cheaper approach is also correct on the Tuesday.

Exam Room · AI Practitioner

Pop Quiz: Where the Embeddings Go

Subscribers and catalogue already live in Amazon Aurora PostgreSQL. The team wants semantic search over product descriptions and has said plainly that it will not run, secure and pay for a second data store. Which store holds the vectors?

Exam Room · AI Practitioner

Pop Quiz: Examples or Reasoning

A refund prompt applies a tiered policy with a pro-rata rule and a cap, and gets one in five wrong. Eight worked examples have not moved the error rate. What changes next?

Exam Room · AI Practitioner

Pop Quiz: Naming the Prompt Attack

A supplier uploads a spec sheet with one odd sentence buried in it, and a week later the assistant reads the internal pricing table out to anyone who asks. Four named risks describe part of it. Only one names the mechanism.

Exam Room · AI Practitioner

Pop Quiz: ROUGE or BLEU

Two candidate models, 500 incident reports, and a human-written summary of each already in a spreadsheet. The team wants a number per model by Friday. Which automatic metric scores a summariser?

Exam Room · AI Practitioner

Pop Quiz: Distillation or Continued Pre-Training

A support assistant on a flagship model answers well. At forty thousand calls a day it costs too much per call and replies too slowly. Which training route fixes that?

Exam Room · AI Practitioner

Pop Quiz: One Prompt or an Agent

A nightly report runs the same four steps it ran last night, and somebody has proposed building it as an agent. When every step is known before the job starts, an agent loop only adds model calls.

Exam Room · AI Practitioner

Pop Quiz: The Metric the Business Reads

A self-service assistant is up for renewal with ROUGE-L at 0.51 and containment up 30%. Repeat contacts within a day are also up, and the agents taking the handovers say customers arrive further from a resolution. Which number settles it?

Exam Room · AI Practitioner

Pop Quiz: Capping the Response

A product-description generator writes two or three paragraphs more than the page has room for, and resends the same three thousand tokens of brand guidelines on every call. Two settings deal with both halves of that bill, and neither of them is temperature.

Exam Room · AI Practitioner

Pop Quiz: Rolling Back a Prompt

Five services share one piece of customer-facing wording. Somebody edited it last Tuesday, the answers got worse, and nobody kept a copy of what it used to say.

Exam Room · AI Practitioner

Pop Quiz: The Smaller Model and the Emissions Target

A published emissions target, a summarisation feature to build, and a sign-off form with a box for the energy cost. Five answers, and one of them reduces the compute rather than accounting for it.

Exam Room · AI Practitioner

Pop Quiz: Who Owns the Generated Image

Marketing wants generated images and copy on the public site. Legal wants to know what happens when a photographer says the picture is theirs. Four of the five controls on the table answer a different worry entirely.

Exam Room · AI Practitioner

Pop Quiz: Imbalanced Data or Bad Labels

A fraud training set holds 200,000 legitimate transactions and 1,100 fraudulent ones, and two annotators disagreed on one label in six. Rebalancing fixes one defect of the two.

Exam Room · AI Practitioner

Pop Quiz: Overfitting, Underfitting, or Too Few Examples

96 per cent on training data, 94 per cent on validation, 71 per cent for one group of customers. Three of the four repairs on offer are aimed at a fault the first two numbers rule out.

Exam Room · AI Practitioner

Pop Quiz: Interpretable or Explainable

Every declined application has to arrive with the reasons it was declined. Four candidate models, and only two can supply them.

Exam Room · AI Practitioner

Pop Quiz: Model Card, Service Card, or Artifact

An auditor asks three things in one meeting: what your own model is for, what AWS documented about the managed service you call, and proof that AWS holds SOC 2. Three artefacts, and the tempting answer merges two of them.

Exam Room · AI Practitioner

Pop Quiz: The Thumbs-Down Button

A customer reads an answer from the support assistant, knows it is wrong about her own account, and has nowhere to say so. Four of the five fixes on the table aim at somebody who will never meet her.

Exam Room · AI Practitioner

Pop Quiz: What a VPC Endpoint Actually Does

Told to take Bedrock calls off the public service endpoint, the team added an interface VPC endpoint. The register now says the traffic is encrypted and access is restricted. One of those entries is wrong.

Exam Room · AI Practitioner

Pop Quiz: Config, CloudTrail, Inspector, or Artifact

An assessor wants proof that a bucket has blocked public access every day for a year. Several governance services sound like they answer that, and the tempting one records only the moments somebody touched the setting.

Exam Room · AI Practitioner

Pop Quiz: When a Model Says It Is Confident

The assistant ends every answer with a confidence figure, so the team plans to auto-approve anything at 95% or above. It reads like a sensible gate. It is measuring the wrong thing.

Exam Room · AI Practitioner

Pop Quiz: Does Picking a Region Settle Residency

The assistant runs in Sydney because the rule says customer data stays in Australia. Legal wants that confirmed in writing. Three things can still move the data or outlive it, none of them in the Region setting.

Exam Room · AI Practitioner

Pop Quiz: Lineage, Catalogue, or Model Card

One audit email asks four questions, and each starts with where did this come from. Four artefacts answer them, and the tempting move is to reach for the Model Card four times because it sounds most official.

Exam Room · AI Practitioner

Pop Quiz: Access Control or Integrity

Three requirements land on one corpus: analysts must not see another country's customers, documents must survive seven years unaltered, and thin batches must not load. The first is the one people misfile.

Exam Room · AI Practitioner

Pop Quiz: Which Log Records What Was Said

A customer quotes an answer the assistant supposedly gave, and the team is sure it never produced that text. Four logs are running and one holds the words.

Exam Room · AI Practitioner

Flash Card: Choosing a Foundation Model

The factors that decide which foundation model a feature runs on: model type and modality, capabilities against the performance requirements, context window, latency and cost, and the constraints that rule candidates out first.

Exam Room · AI Practitioner

Flash Card: What GenAI Gets Used For

The generative-AI use cases Domain 2 lists, and the AWS service that usually runs each one.

Exam Room · AI Practitioner

Flash Card: Amazon Bedrock AgentCore

The managed platform for running agents in production on AWS: a serverless runtime that hosts whatever framework built the agent, plus memory, a tool gateway, scoped identity and a trace of every call.

Exam Room · AI Practitioner

Flash Card: Strands Agents

The open-source AWS framework for writing an agent in code. Where the agent then runs, a laptop or a managed runtime, is a separate decision.

Exam Room · AI Practitioner

Flash Card: Vector Storage on AWS

Several AWS services now store embeddings and answer a similarity query. Learn them by the phrase in a scenario that gives each one away.

Exam Room · AI Practitioner

Flash Card: Temperature, Top-P and Max Tokens

The four sliders on a model call that change the answer without changing the model, and what each one is actually for.

Exam Room · AI Practitioner

Flash Card: Amazon Bedrock Guardrails

A configurable safety layer that checks prompts on the way in and responses on the way out, outside the model.

Exam Room · AI Practitioner

Flash Card: AWS AI Service Cards

AWS's own transparency document for a managed AI service, written by AWS about AWS. Learn it by who holds the pen, because that is what separates it from the card you write yourself.

Exam Room · AI Practitioner

Flash Card: The Generative AI Security Scoping Matrix

AWS places a generative AI use case in one of five scopes by how much of the model and training data you own. The scope decides which controls are yours.

Exam Room · AI Practitioner

Flash Card: AgentCore Identity and Policy

Two names on agent security that answer different questions: who is asking, and what that caller may do once it is let in. Neither replaces IAM.

Exam Room · AI Practitioner

Flash Card: What Bedrock Does With Your Prompts

Almost everyone new to hosted models assumes the service trains on what they type. It does not. Five lines settle the rest: training use, encryption, the Region that processes it, where it is written down, and what a fine-tune leaves behind.

Exam Room · AI Practitioner

A Domain-by-Domain Checklist for the AI Practitioner Exam

The AI Practitioner certification scores five domains, weighted from 28% down to 14%. This is the whole AI Practitioner track on one page, sorted into those five domains: the cheat sheet that anchors each one, the decisions to work through, and the quizzes and cards to drill. Tick your way down, and wherever a line still feels shaky, that is where the next hour goes.

Exam Room · Advanced Generative AI Developer

Deciding Which GenAI Security Controls Are Yours to Own

Four generative-AI uses inside one retailer, and every security conversation about them stalls on the same question: which controls do we run, and which are we trusting somebody else to run? That answer is not a property of the vendor or the service tier. It follows from how much of the stack you built, which is what the Generative AI Security Scoping Matrix is for.

Exam Room · Advanced Generative AI Developer

The Demo That Had to Justify Itself

A developer spent a week wiring a demo that reads a mine-site callout and proposes the parts to load on the truck. Thirty inputs, twenty-seven that looked right, one impressed executive, and a funding proposal for AUD$310,000. What the demo has not produced is a number anyone could have failed, a sample nobody chose, or a measured token cost. Turning it into a defensible go or no-go is a week of different work.

Exam Room · Advanced Generative AI Developer

Getting a GenAI Feature to the People Who Use It

One Bedrock-backed summariser, two teams, two different asks. One needs a chat interface in front of depot staff inside a fortnight and has no backend capacity. The other needs the request and response contract settled before either side writes a handler, because the engineers building against it work for another company. AWS Amplify answers the first and an OpenAPI document answers the second, and using either one for both jobs is how a fortnight becomes a quarter.

Exam Room · Cloud Practitioner

Cheat Sheet: Cloud Concepts

A final-review sheet for CLF-C02 Domain 1: the value proposition of the cloud, the six Well-Architected pillars and what separates them, the Cloud Adoption Framework, the seven migration strategies, and the economics that turn a data centre into a monthly bill.

Exam Room · Cloud Practitioner

Forty Applications and a Lease That Ends in March

A regional insurer will not be renewing its data centre lease, which gives forty applications about seven months to be somewhere else. Rewriting all of them is not available. Lifting all of them wastes the one chance to fix the worst of them. Sorting forty applications into the right seven buckets is the actual work, and the sorting rule is not technical.

Exam Room · Cloud Practitioner

The Servers We Sized for One Tuesday in November

An online retailer runs at 8% CPU for eleven months so that one Tuesday in November does not fall over. The finance director wants to know why the cloud would be cheaper when the sticker price per hour looks worse. Answering that means separating what is fixed from what is variable, and counting the costs an invoice never shows.

Exam Room · Cloud Practitioner

Flash Card: The AWS Well-Architected Framework

Six pillars, a named set of design principles under each, lenses for particular workload types, and a console tool that turns the questions into an improvement plan. Reach for it whenever a scenario describes a property of a design.

Exam Room · Cloud Practitioner

Pop Quiz: Elasticity, Scalability and Agility

A ticketing platform adds capacity when a sale opens and releases it two hours later, with no human involved, and the charges stop when the capacity goes. Which benefit of the cloud is that?

Exam Room · Cloud Practitioner

Cheat Sheet: AWS Security and Compliance

A final-review sheet for CLF-C02 Domain 2: the shared responsibility model and how it shifts by service, identity and access management, the root user, encryption, and the services that detect, audit and report. Thirty per cent of the scored content, condensed.

Exam Room · Cloud Practitioner

The Patch That Nobody Owned

A critical vulnerability lands in a library that appears in five places across one estate: on EC2 instances, inside an RDS engine, under a Lambda function, in a Fargate task image, and in a third-party appliance from AWS Marketplace. Most of them need the company to act by Friday. One does not, and one splits in two depending on how the library got there. The shared responsibility model is what tells them apart.

Exam Room · Cloud Practitioner

One Login and Eleven People

A startup has grown to eleven engineers and still signs in to AWS with the account root credentials, kept in a shared password manager. An access key from that identity is in a build script. Nobody can say who deployed anything, and the first audit is in six weeks. Untangling it is a lesson in what identities exist and what each one is for.

Exam Room · Cloud Practitioner

Four Findings and Four Different Services

A health-tech company has four security questions on one page: is anything malicious happening, do our instances carry known vulnerabilities, is there patient data in the wrong bucket, and are we drifting off the standard we told the auditor we follow. Four questions, four different services, and the way to tell them apart is what each one reads.

Exam Room · Cloud Practitioner

Two Encryption Choices and One Service Out of Scope

A Perth company moving a case-management platform for a Commonwealth agency into AWS has three things to settle before the assessor arrives: encrypt 14TB of scanned records properly, answer a finance director who thinks the locked cage in the leased data centre was safer, and confirm that every service in the design sits inside the compliance programme the platform is assessed against. One of them does not.

Exam Room · Cloud Practitioner

Flash Card: The Shared Responsibility Model

AWS secures the cloud; the customer secures what they put in it. The line between the two moves with the service, and three responsibilities never move at all. Security and Compliance carries 30% of the scored content, and this model sits underneath all of it.

Exam Room · Cloud Practitioner

Pop Quiz: What Only the Root User Can Do

Four of these five tasks an AdministratorAccess identity can do. One of them nobody can do without signing in as the root user, and it is not the one most study notes name.

Exam Room · Cloud Practitioner

Pop Quiz: CloudTrail, Config and CloudWatch

A security group that should allow port 443 from one address range is found allowing 0.0.0.0/0. Two questions follow: what does this security group look like now compared to last month, and who made the call that changed it? They are answered by two different services.

Exam Room · Cloud Practitioner

Cheat Sheet: Cloud Technology and Services

A final-review sheet for CLF-C02 Domain 3: ways to reach AWS, the global infrastructure, and the compute, storage, database, network, AI/ML, analytics, integration and developer services, with the one-line job each of them does. The largest domain, and the one that is mostly recall.

Exam Room · Cloud Practitioner

Three Copies in One Building

A booking platform runs three web servers, a database and its standby, and all of it sits in one Availability Zone. The team believes it is redundant. A four-hour power event in that zone proves otherwise. Fixing it means being precise about what a Region is, what an Availability Zone is, and which problem an edge location solves.

Exam Room · Cloud Practitioner

Five Ways to Run the Same Container

A media company has four workloads and one team: an image resizer triggered by uploads, a licensed transcoding engine, a steady web API, and an overnight rendering job. Every one of them could technically run on an EC2 instance. Only one of them should, and the reason is different in each case.

Exam Room · Cloud Practitioner

Six Petabytes and Nobody Knows Which Drawer

A genomics institute has six petabytes spread across five kinds of data: sequencer output being processed right now, a shared analysis workspace, results served to a public portal, twenty years of raw reads kept for regulatory reasons, and a scratch area rebuilt nightly. One storage service is wrong for all five, and the discriminator is how the data is reached rather than how big it is.

Exam Room · Cloud Practitioner

A Table That Fits and One That Does Not

A ride-hailing company runs everything in one PostgreSQL database: bookings, driver location updates arriving forty thousand a second, session tokens, and the analytics queries that lock the whole thing up every morning at nine. Four workloads in one engine is the problem. Splitting them is a lesson in what each database service is actually for.

Exam Room · Cloud Practitioner

A Subnet, a Gateway and a Route Nobody Wrote

A charity moves its donation platform into a VPC and nothing works: the web tier is unreachable, the application servers cannot fetch patches, and the nightly backup to S3 has been failing since the move. The obvious fix to the first of those would put a database full of donor records on a public path. Four symptoms, four different pieces of VPC plumbing, and each one is a component with a specific job.

Exam Room · Cloud Practitioner

Flash Card: AI, Machine Learning and Analytics Services

Two catalogues to know by recognition rather than depth: the AI services that each do one task, and the analytics services that move, query and visualise data. Learn them by the job each one does in a sentence.

Exam Room · Cloud Practitioner

Flash Card: Integration, Developer and End-User Services

The rest of the catalogue: messaging and events, developer tools, business applications, virtual desktops, frontend hosting and IoT. Each one is recognised by the single job it does.

Exam Room · Cloud Practitioner

Cheat Sheet: Billing, Pricing and Support

A final-review sheet for CLF-C02 Domain 4: the compute purchasing options and when each fits, storage and data transfer charges, the billing and cost management tools, AWS Organizations and cost allocation tags, the Support plans and their response times, and the partner and technical resources.

Exam Room · Cloud Practitioner

A Question Nobody on the Team Could Answer

An eleven-person company in Adelaide spends an afternoon on a problem nobody can solve, and works out halfway through that the real difficulty is not knowing who to ask. Five more questions arrive over the next month, and each one has a different route: a support case, a page AWS has already written, somebody else's software, or a partner with hands on the work.

Exam Room · Cloud Practitioner

The Same Instance at Four Prices

A market research firm runs 180 instances, all of them On-Demand, and finance wants the bill down without anything being switched off. The estate divides into four groups with completely different demand shapes, and the same instance type is available at four different rates depending on what you are prepared to commit to. What each group can honestly commit to decides the rate it gets.

Exam Room · Cloud Practitioner

One Bill, Nine Teams and Nobody to Ask

A software company with nine product teams gets one AWS invoice a month, and nobody can say which team spent what. A finance business partner wants chargeback, a team lead wants warning before they overspend, and the platform lead wants a support response within the hour rather than the next business day. Three different tools, and one organisational change underneath all of them.

Exam Room · Cloud Practitioner

A Domain-by-Domain Checklist for the Cloud Practitioner Exam

The Cloud Practitioner paper scores four domains, from Cloud Technology and Services at 34% down to Billing, Pricing and Support at 12%. This is the whole track on one page, sorted into those four: the cheat sheet that anchors each domain, the situations to work through, and the quizzes and cards to drill. Tick your way down, and wherever a line still feels shaky, that is the next half hour of revision.

Exam Room · AI Business Strategist

Forty AI Tools Nobody Approved

An expenses review at a 1,400-person professional services firm turns up forty-one AI subscriptions bought in a year, most on personal cards, several holding client files. A blanket ban was issued eight months ago and quietly ignored. A ban, a closed allow-list and a published three-state register all look like governance from the outside; only one of them gets people to tell you what they are actually running.

Exam Room · AI Business Strategist

Four Asks and One Budget

Four functions at a manufacturer each arrive with an AI ask, and the board has money for one of them. None can be priced until somebody names the capability each one actually needs, and naming it changes three of the four answers.