From Prompts to Production Your AI Practitioner Prep Starts Now

Take on 25 practice questions across five AI Practitioner domains
Start Free Practice
No Credit Card Required
HOW TO PREPARE

Everything You Need to Know First

Warm-up topics before the test.
  • Foundational AWS Knowledge: Familiarity with core services like EC2, S3, Amazon Bedrock, Lambda, and IAM, plus the shared responsibility model.
  • AI and ML Basics: General awareness of machine learning, generative AI, and foundation model concepts.
  • Responsible AI Awareness: Basic understanding of bias, fairness, transparency, and data privacy considerations.
PRACTICE TEST

Put Your Knowledge to the Test

25 multiple-choice questions, including ones built around command output
A retrieved document tells an AI assistant to ignore its instructions and send confidential data to an external destination. What kind of threat is this?
A routine data-retention policy.
A normal instruction from a trusted application administrator.
Model quantization.
Indirect prompt injection.
D
Malicious instructions can arrive through external content such as retrieved documents. Treat retrieved text as untrusted data and enforce access and action controls outside model instructions alone.
Security, Compliance, and Governance for AI Solutions
25
Select One
An auditor asks which IAM identity made a supported AWS management API call affecting the AI environment. Which service records this type of account activity?
Amazon Translate.
Amazon Route 53 routing policies.
AWS CloudTrail.
Amazon Polly.
C
CloudTrail records supported AWS account/API activity and identity details. This does not mean every model prompt and response is recorded automatically; model invocation content logging is a separate concern.
Security, Compliance, and Governance for AI Solutions
24
Select One
An application should invoke only the approved Bedrock models needed for its task. Which access-control approach follows least privilege?
Make all stored prompts public so permissions are unnecessary.
Use an IAM role and policies scoped to the required model resources and invocation actions.
Share the account root credentials across the application team.
Give the application AdministratorAccess.
B
IAM roles and scoped policies restrict AWS API access to what the application needs. Guardrails and model settings do not replace IAM authorization.
Security, Compliance, and Governance for AI Solutions
23
Select One
A team wants to document a model's intended use, evaluation results, and limitations for stakeholders. Which SageMaker feature is designed for this?
An Auto Scaling launch template.
An S3 multipart upload.
Amazon SageMaker Model Cards.
A VPC route table.
C
Model Cards consolidate model documentation for transparency and governance. Documentation does not itself certify that the model is unbiased or appropriate for every use.
Guidelines for Responsible AI
22
Select One
An application needs configurable filtering for harmful content and specified denied topics around foundation-model interactions. Which Bedrock capability directly supports this?
Amazon Bedrock Guardrails.
A Route 53 hosted zone.
An S3 lifecycle policy.
A larger maximum output-token setting.
A
Guardrails provide configurable content and topic protections. They reduce specified risks but are not a guarantee that every harmful or inaccurate response will be blocked.
Guidelines for Responsible AI
21
Select One
A company uses an AI system to draft recommendations with significant consequences for customers. Which practice supports meaningful human oversight?
Give reviewers enough context, limitations, and authority to reject or correct recommendations.
Require reviewers to accept every model output.
Hide known limitations from reviewers.
Treat a model's confident tone as a substitute for review.
A
Meaningful oversight requires informed judgment and the authority to intervene - which is why qualified people should review consequential recommendations before action is taken, not merely rubber-stamp them. A nominal reviewer who cannot question or change the result does not provide that control.
Guidelines for Responsible AI
20
Select One
A model performs well overall but has a much higher error rate for one demographic group. Which next step best supports responsible evaluation?
Remove all evaluation because the overall score is high.
Report only aggregate accuracy and discard subgroup results.
Assume equal sample counts automatically guarantee fairness.
Measure performance by relevant subgroups and investigate representation, labels, and sources of bias.
D
Aggregate scores can hide unequal outcomes. Subgroup analysis and investigation can reveal bias, but no single metric or sampling rule guarantees fairness.
Guidelines for Responsible AI
19
Select One
A simple vector-based RAG system has already indexed its documents. What is the correct dependency order of these request-time steps: generate an answer from the augmented prompt; convert the user's query into a compatible embedding; retrieve relevant document chunks using that query embedding; construct a prompt containing the user's question and retrieved context?
Retrieve chunks → Convert query to embedding → Generate answer → Construct prompt
Convert query to embedding → Retrieve relevant chunks → Construct augmented prompt → Generate answer
Generate answer → Construct prompt → Retrieve chunks → Convert query to embedding
Construct prompt → Convert query to embedding → Generate answer → Retrieve chunks
B
In this vector-based flow, the query embedding is needed for retrieval, retrieved chunks are needed to construct the augmented prompt, and generation uses that prompt. Other RAG architectures can include additional steps.
Applications of Foundation Models
18
Select One
An assistant must interpret a request, look up an order through an API, and invoke an approved action when appropriate. Which capability is most relevant?
A storage lifecycle rule.
An agent configured with tools or action groups for the permitted operations.
An embedding model with no tool integration.
A tokenizer operating by itself.
B
Agents can orchestrate multistep tasks and invoke configured tools. Tool permissions, validation, and appropriate approval controls remain part of the application design.
Applications of Foundation Models
17
Select One
A team has a curated set of domain-specific input/output examples and wants to adapt a supported model's parameters to follow that task pattern. Which approach directly does this?
RAG alone without model training.
Adding an IAM role.
Fine-tuning.
Changing an S3 bucket name.
C
Fine-tuning trains a pretrained model further on task data. RAG supplies information at inference time and does not by itself update model parameters.
Applications of Foundation Models
16
Select One
A supported text model exposes temperature as a sampling parameter. With other settings unchanged, what is the usual effect of lowering it?
Every answer becomes guaranteed correct and identical.
The model's context window automatically doubles.
The model gains access to current internet facts.
Responses tend to be less random and favor more probable token choices.
D
Lower temperature generally reduces sampling randomness. It does not add knowledge or guarantee correctness or complete determinism across all systems.
Applications of Foundation Models
15
Select One
A prompt includes three examples of customer messages paired with the desired classification, followed by a new message to classify. Which prompting technique is this?
Zero-shot prompting.
Few-shot prompting.
Pretraining from scratch.
Model-weight quantization.
B
Few-shot prompting provides examples in the prompt. It guides inference without necessarily changing model parameters.
Applications of Foundation Models
14
Select One
SCENARIO (continued): Before releasing the retailer's assistant, the team must assess whether its answers are correct and supported by the cited policies. Which evaluation best addresses this requirement?
Treat the presence of any citation as proof that the answer is correct.
Check only whether every answer has more than 100 words.
Select the model that generates the most tokens per second without inspecting answers.
Test representative support questions and have reviewers check answer correctness and source support against approved policies.
D
Representative task evaluation and source checks test the required behavior. A citation can be irrelevant or fail to support a claim; speed and answer length alone do not establish correctness.
Applications of Foundation Models
13
Select One
SCENARIO: A retailer is building a customer-support assistant. Product policies change weekly. Approved policy documents are searchable. The assistant must use current policy information and identify supporting sources. Which approach best supplies current knowledge without retraining the model after every policy change?
Retrieve relevant approved documents at request time and provide them as context for generation.
Fine-tune the model once and stop updating its information.
Rely only on the model's pretraining knowledge.
Increase temperature so the model can invent missing policy details.
A
RAG retrieves current external information for the prompt rather than relying exclusively on model weights. Keeping the index current and checking citations remain necessary.
Applications of Foundation Models
12
Select One
A selected model's on-demand text inference is priced separately by input and output tokens. Which change can reduce token-based charges while preserving the application's required quality?
Add unrelated documents to every prompt.
Increase the output limit and force every answer to fill it.
Retry every successful response automatically.
Reduce unnecessarily verbose generated responses and choose an appropriate output limit.
D
Less unnecessary input and output can reduce token-based usage: trimming repeated context out of prompts lowers input tokens, while reducing unnecessarily verbose responses and choosing an appropriate output limit lowers output tokens. An output limit caps generation but does not guarantee actual length; evaluate quality when changing prompts or limits.
Fundamentals of Generative AI
11
Select One
A team wants managed API access to a choice of foundation models without operating model-serving servers. Which AWS service is the most direct fit?
Amazon Bedrock.
AWS CloudTrail.
Amazon EC2 alone with a self-managed model server.
Amazon Route 53.
A
Bedrock provides managed foundation-model access and related application capabilities. Self-managed EC2 hosting requires operating the serving infrastructure.
Fundamentals of Generative AI
10
Select One
A language model confidently cites a research paper that does not exist. Which limitation does this illustrate?
Encryption at rest.
Autoscaling.
Hallucination.
Deterministic validation.
C
A hallucination is generated content that is incorrect or unsupported. Fluent wording and confident tone are not evidence that a claim is true.
Fundamentals of Generative AI
9
Select One
A semantic search application converts text into numerical vectors so it can compare meaning. What are these vectors commonly called?
Embeddings.
Database passwords.
IAM policies.
Network routes.
A
Embeddings represent information numerically in a vector space. A similarity search can compare these representations; the representation is not an access-control mechanism.
Fundamentals of Generative AI
8
Select One
Which statement about tokens in a text foundation model is accurate?
A token can represent a word, part of a word, or another text unit, depending on the tokenizer.
Every token always corresponds to exactly one English word.
Tokenization is identical across every foundation model.
A token is always a complete paragraph.
A
Token boundaries depend on the model's tokenizer. Word count and token count are not interchangeable.
Fundamentals of Generative AI
7
Select One
Which task is a direct example of generative AI?
Sorting rows numerically by an existing price column.
Checking whether a file's checksum matches an expected value.
Applying a fixed tax percentage with a deterministic formula.
Drafting an original product description from supplied product details.
D
Generative AI creates content from learned patterns and supplied context. Sorting, checksum comparison, and a fixed arithmetic rule do not require content generation.
Fundamentals of Generative AI
6
Select One
Which AWS service converts recorded speech into text?
Amazon Translate
Amazon Transcribe
Amazon Polly
Amazon Comprehend
B
Transcribe performs speech-to-text, Polly performs text-to-speech, and Translate performs language translation. These are distinct input/output tasks.
Fundamentals of AI and ML
5
Select One
After deployment, incoming data begins to differ from the model's baseline data. Which AWS capability is designed to help detect data-quality drift for supported SageMaker deployments?
Amazon Route 53 health checks.
Amazon Polly.
Amazon SageMaker Model Monitor.
AWS CloudFormation drift detection alone.
C
Model Monitor checks data and model-quality conditions against configured baselines and constraints. It can alert on violations; detection does not automatically solve the underlying modeling problem.
Fundamentals of AI and ML
4
Select One
A model must predict a property's sale price as a numerical amount from its characteristics. Which problem type is this?
Regression.
Machine translation.
Clustering only.
Multiclass classification.
A
A continuous numeric target is a regression problem. Classification assigns categories rather than directly predicting an amount.
Fundamentals of AI and ML
3
Select One
A retailer wants to discover groups of customers with similar purchase behavior. There are no predefined group labels. Which technique is most appropriate?
Speech recognition.
Clustering.
Supervised binary classification.
Text-to-speech synthesis.
B
Clustering is an unsupervised approach for discovering group structure in unlabeled data. It does not require predefined customer-segment targets.
Fundamentals of AI and ML
2
Select One
A company has historical support tickets labeled billing, technical, or account. It wants a model to assign one of these labels to a new ticket. Which approach fits?
Unsupervised clustering without labels.
Supervised multiclass classification.
Reinforcement learning driven only by game rewards.
Regression predicting a continuous number.
B
The target is one of several known categories, and labeled examples are available. Regression predicts numeric values; clustering discovers groups without using these target labels.
Fundamentals of AI and ML
1
Select One
DOMAIN Question 01/00

0 of 0 answered

A Course Built Around Your Needs

AWS Certified AI Practitioner

Beginner
Associate
Professional
08.20
Hours
Play icon
Master AWS Certified AI Practitioner concepts through hands-on modules and practical AWS applications.
Michael Forrester
Details
WHY HANDS-ON PRACTICE

Knowledge, Earned by Doing

The AI Practitioner exam rewards applied AI judgment, not memorized definitions.

Build Real Exam Speed

‍The exam gives you about 83 seconds per question on average. Repeated practice trains you to spot the right AWS service fast, so you're not re-reading scenarios under time pressure.

Know Your Weak Domains

‍Review every question you missed and spot the topics that need more work. Better to find them here than on exam day.

‍

Realistic Scenario Questions

‍Exam-style questions, in the multiple-choice format you'll see on the real test.

‍

Reduce Exam-Day Anxiety

‍Familiarity with multiple-choice format removes surprises, so nerves don't cost you marks.

‍

Reinforce Through Repetition

‍Spaced repetition on frequently tested concepts, like Bedrock, prompt engineering, and model evaluation, moves them from look it up to just know it.

Benchmark Against a Passing Score

‍Compare your score with the passing threshold and identify topics to review before exam-day.

HEAR FROM OUR LEARNERS

Real Learners. Real Results

Alhamdulillah 🙏! Thrilled to announce that I am now officially a Certified Kubernetes Administrator (CKA)! While Kubernetes, cluster orchestration, and container security are already central to my daily DevSecOps work, earning this certification is a great validation of those hands-on skills.Thanks to Mumshad Mannambeth and the KodeKloud team—their course and hands-on labs helped me a lot in preparing for this exam.The learning never stops! 🚀

Read Full Post
Imtiaz Ali
DevOps | SRE | Multi-Cloud | DevSecOps

Freshly certified: Certified Kubernetes Administrator (CKA)! 🚀Passing this hands-on, live-terminal exam was a great challenge. Paired with my AWS Certified DevOps Engineer – Professional, it reinforces my focus on bridging enterprise multi-account cloud governance with production-grade Kubernetes runtime operations.A special shoutout and big thank you to Mumshad Mannambeth, the KodeKloud team, and the Cloud Native Computing Foundation (CNCF) and Kubernetes (Official) community. The Certified Kubernetes Administrator (CKA) course was hands-down the best preparation for the live terminal environment! 🛠️Next stop: continuing to build resilient, developer-first platform foundations!

Read Full Post
Gaurav Dubey
Cloud Architect | Platform Engineer | AWS DevOps Pro | CKA

Cherishing one of my dreams - CKS Certified! ✨Last year, in October, when I cracked my CKA, my colleagues asked, “When is CKS?” 😂And I was like… bro. 😅Because it felt so much bigger. I also knew it wasn’t just Kubernetes anymore, but Kubernetes with a special layer of security - yeah, tightened Kubernetes.Back then, I used to look up to CKS like a child gazing at one of the stars in the sky.⭐Now, fast forward to today - I cleared it last week and officially became a Certified Kubernetes Security Specialist (CKS), thanks to the Cloud Native Computing Foundation (CNCF), The Linux Foundation and my company Aivar Innovations. 🎉That moment felt so special, and I’m incredibly grateful to all my colleagues, friends, and family for all the support. ❤️A mention of the resources I used: KodeKloud by Mumshad Mannambeth and Kim Wüstkamp’s CKS courses. I also had my colleagues supporting me with my doubts and anxieties before the exam. 😂Looking forward to applying everything I’ve learned in my work and continuing to grow in the Cloud Native and Kubernetes space. 🚀So, cheers to everyone who dreams, aspires, and inspires. Keep going and keep up the good work! 🥂And as I always say - God, keep it coming!! ✨

Read Full Post
Rima Sidique
DevOps Engineer | AI & DevOps Enthusiast

I've successfully passed CKA certification. Shout-out to Mumshad Mannambeth for his great CKA course and labs and to Omkar Shelke Killercoda CKA scenarios. Links: - https://lnkd.in/eZPVHJeR- https://lnkd.in/eEVyZdaZ

Read Full Post
Gabriel Rodríguez
DevSecOps Engineer

🎉 The Kubernetes journey continues… and this time, happy to say I’m officially a Kubestronaut! 🧑🚀🚀🎖️ 1 out of 242 in France and 1 out of 4217 worldwide!⏳✈️ It’s been quite a journey. From my early days working with Kubernetes, thanks to the opportunity Prof. Timur Friedman gave me to contribute to the EdgeNet project, to securing my internship as a DevOps Engineer at Sonos, Inc. — and fast-forward five years, here I am today, a Senior DevOps Engineer at Sonos, Inc. working extensively with Kubernetes everyday.💌 A special thank you to Jose Alfredo Alvarez Aldana for your support, guidance, and for everything you taught me along the way. I was really lucky to have the opportunity to work alongside you early in my journey. 🙏💯 A big thank you to Mumshad Mannambeth for building KodeKloud! The platform has been incredibly helpful throughout this journey, both for learning and preparing for the exams.And of course, shout out to all the people who supported, challenged, and inspired me along the way. ❤️Never give up 💪Keep smiling 😃

Read Full Post
Rima Sidique
DevOps Engineer | AI & DevOps Enthusiast

Alhamdulillah 🙏! Thrilled to announce that I am now officially a Certified Kubernetes Administrator (CKA)! While Kubernetes, cluster orchestration, and container security are already central to my daily DevSecOps work, earning this certification is a great validation of those hands-on skills.Thanks to Mumshad Mannambeth and the KodeKloud team—their course and hands-on labs helped me a lot in preparing for this exam.The learning never stops! 🚀

Read Full Post
Imtiaz Ali
DevOps | SRE | Multi-Cloud | DevSecOps

Freshly certified: Certified Kubernetes Administrator (CKA)! 🚀Passing this hands-on, live-terminal exam was a great challenge. Paired with my AWS Certified DevOps Engineer – Professional, it reinforces my focus on bridging enterprise multi-account cloud governance with production-grade Kubernetes runtime operations.A special shoutout and big thank you to Mumshad Mannambeth, the KodeKloud team, and the Cloud Native Computing Foundation (CNCF) and Kubernetes (Official) community. The Certified Kubernetes Administrator (CKA) course was hands-down the best preparation for the live terminal environment! 🛠️Next stop: continuing to build resilient, developer-first platform foundations!

Read Full Post
Gaurav Dubey
Cloud Architect | Platform Engineer | AWS DevOps Pro | CKA

Cherishing one of my dreams - CKS Certified! ✨Last year, in October, when I cracked my CKA, my colleagues asked, “When is CKS?” 😂And I was like… bro. 😅Because it felt so much bigger. I also knew it wasn’t just Kubernetes anymore, but Kubernetes with a special layer of security - yeah, tightened Kubernetes.Back then, I used to look up to CKS like a child gazing at one of the stars in the sky.⭐Now, fast forward to today - I cleared it last week and officially became a Certified Kubernetes Security Specialist (CKS), thanks to the Cloud Native Computing Foundation (CNCF), The Linux Foundation and my company Aivar Innovations. 🎉That moment felt so special, and I’m incredibly grateful to all my colleagues, friends, and family for all the support. ❤️A mention of the resources I used: KodeKloud by Mumshad Mannambeth and Kim Wüstkamp’s CKS courses. I also had my colleagues supporting me with my doubts and anxieties before the exam. 😂Looking forward to applying everything I’ve learned in my work and continuing to grow in the Cloud Native and Kubernetes space. 🚀So, cheers to everyone who dreams, aspires, and inspires. Keep going and keep up the good work! 🥂And as I always say - God, keep it coming!! ✨

Read Full Post
Rima Sidique
DevOps Engineer | AI & DevOps Enthusiast

I've successfully passed CKA certification. Shout-out to Mumshad Mannambeth for his great CKA course and labs and to Omkar Shelke Killercoda CKA scenarios. Links: - https://lnkd.in/eZPVHJeR- https://lnkd.in/eEVyZdaZ

Read Full Post
Gabriel Rodríguez
DevSecOps Engineer

🎉 The Kubernetes journey continues… and this time, happy to say I’m officially a Kubestronaut! 🧑🚀🚀🎖️ 1 out of 242 in France and 1 out of 4217 worldwide!⏳✈️ It’s been quite a journey. From my early days working with Kubernetes, thanks to the opportunity Prof. Timur Friedman gave me to contribute to the EdgeNet project, to securing my internship as a DevOps Engineer at Sonos, Inc. — and fast-forward five years, here I am today, a Senior DevOps Engineer at Sonos, Inc. working extensively with Kubernetes everyday.💌 A special thank you to Jose Alfredo Alvarez Aldana for your support, guidance, and for everything you taught me along the way. I was really lucky to have the opportunity to work alongside you early in my journey. 🙏💯 A big thank you to Mumshad Mannambeth for building KodeKloud! The platform has been incredibly helpful throughout this journey, both for learning and preparing for the exams.And of course, shout out to all the people who supported, challenged, and inspired me along the way. ❤️Never give up 💪Keep smiling 😃

Read Full Post
Rima Sidique
DevOps Engineer | AI & DevOps Enthusiast
FAQs

Frequently Asked Questions

Is this practice test free ? 

Yes! This practice test is completely free. You can take it as many times as you like to practice and assess your knowledge.

Do I need to install anything ? 

No. You don’t need to install anything. The practice test is completely browser-based, so you can start practicing directly online. 

What if I'm new to this domain?

That's completely fine , mock tests aren't just for the final week before your exam. Taking one early gives you a baseline: you'll see which topics you already grasp, which ones you've never encountered, and how the real exam frames its questions. Every attempt comes with detailed explanations for each answer, so even the questions you get wrong become a study guide. We'd suggest starting with one full mock test, reviewing the explanations carefully, then working through the linked learning material before attempting the next one. 

How closely does this mock test match the real exam?

The question style and pattern closely mirror the real certification exam, the way scenarios are framed, the answer options, and the level of reasoning required are all modelled on it. The main difference is length: this mock has 25 sample questions, while the actual exam typically presents 50-60 questions of this type. So treat this as a representative sample rather than a full-length simulation. The style will feel familiar on exam day; the stamina and time pressure of a longer paper is something you'll want to prepare for separately.