AMAADOR ACADEMYCertificatesVerify a certificate

PRO-AWS-AIF

Final check — AWS Certified AI Practitioner (AIF-C01) — Exam-Prep

Answer the 65 questions. You need 80% or more to pass. You can retry as many times as you like.

Study the course →

Question 1A retail company wants software that learns patterns from historical purchase data to predict which products a customer is likely to buy, without a developer writing explicit if-then rules for every product. Which discipline best describes this approach?
Question 2A team labels thousands of emails as 'spam' or 'not spam' and trains a model to predict the label of new emails. What type of machine learning is this?
Question 3A data scientist has customer records with no labels and wants to discover natural groupings of customers based on similar behavior. Which technique is most appropriate?
Question 4What distinguishes deep learning from classic machine learning approaches?
Question 5A model performs extremely well on its training data but poorly on new, unseen data. What is this problem called?
Question 6In a large language model, what is a 'token'?
Question 7A developer wants to find documents semantically similar to a user query, even when they share no exact keywords. Which representation makes this possible?
Question 8Which architecture, based on a self-attention mechanism, underlies most modern large language models?
Question 9A user provides a few input-output examples directly inside the prompt to guide a foundation model before asking it to complete a new case. This prompting technique is known as:
Question 10A prompt engineer adds the instruction 'Let's think step by step' so the model shows intermediate reasoning before its final answer. This technique is called:
Question 11A team gets inconsistent, off-topic answers from a foundation model. Which prompt improvement is MOST likely to help?
Question 12In text generation, what is the typical effect of setting a higher 'temperature' parameter?
Question 13A hiring model was trained mostly on resumes from one demographic group and now systematically scores qualified candidates from other groups lower. This is an example of:
Question 14Which AWS capability helps detect statistical bias in training data and explain feature importance for model predictions?
Question 15A bank must justify to regulators why its credit-scoring model denied a specific application. Which responsible-AI property is being demanded?
Question 16A foundation model confidently produces a factual statement that is completely false and unsupported by any source. This behavior is best described as:
Question 17Which practice best supports transparency about an AI system for its users and stakeholders?
Question 18During the ML lifecycle, which evaluation metric is most appropriate when classes are highly imbalanced (e.g., rare fraud cases) and simply measuring accuracy can be misleading?
Question 19A company wants a foundation model to answer questions using its own private, frequently updated product manuals, without retraining the model. Which approach fits best?
Question 20Which statement correctly describes a foundation model?
Question 21A model that worked well last year now performs poorly because real-world input data has gradually changed over time. What is this phenomenon called?
Question 22To reduce the risk that a generative AI assistant produces harmful or non-compliant content, which control should a team implement?
Question 23A data science team has trained a fraud-detection model. On the test set it correctly flags very few of the actual fraud cases, even though its overall accuracy is 99% because fraud is rare. Which metric should the team prioritize to capture how many true fraud cases the model actually catches?
Question 24A company wants to extract text, form fields, and table data from scanned invoices and tax documents with high accuracy. Which AWS AI service is purpose-built for this?
Question 25A developer wants to build and deploy a generative AI application by calling multiple foundation models (such as Anthropic Claude and Amazon Titan) through a single managed API, without managing any infrastructure. Which AWS service should they use?
Question 26A team is building a model to predict the exact dollar amount of next month's sales based on historical data. What type of machine learning problem is this?
Question 27A contact center wants to automatically transcribe recorded customer phone calls into text so the transcripts can later be analyzed for sentiment. Which AWS service performs the speech-to-text step?
Question 28An organization wants all prompts and responses sent to a foundation model in Amazon Bedrock to be blocked when they contain hate speech, violence, or personally identifiable information. Which Bedrock capability addresses this requirement?
Question 29During training, a model achieves very high accuracy on the training data but performs poorly on new, unseen data. What is this phenomenon called?
Question 30A retailer wants to moderate user-uploaded product images to detect inappropriate or unsafe content and to identify objects in the photos. Which AWS service is most appropriate?
Question 31A company must ensure that data sent to and processed by Amazon Bedrock is not used to train the base foundation models and stays within their AWS environment. Which statement is correct about Amazon Bedrock?
Question 32A team wants a fully managed environment that covers the entire ML lifecycle — labeling, building, training, tuning, and deploying custom models. Which AWS service is the primary platform for this?
Question 33A business wants to control which IAM users and roles are allowed to invoke specific foundation models in Amazon Bedrock. Which AWS service provides this access control?
Question 34A developer needs to build a conversational chatbot with intents, slots, and automatic speech recognition for a customer-service voice and text bot. Which AWS service is designed for this?
Question 35An auditor needs a record of every API call made to Amazon Bedrock and other AWS services, including who made the call and when, for governance and compliance. Which AWS service provides this audit log?
Question 36A team is evaluating a binary classifier and wants a single table that shows the counts of true positives, true negatives, false positives, and false negatives. What is this called?
Question 37A multinational company needs to translate large volumes of product descriptions and support content from English into 15 languages automatically. Which AWS service is purpose-built for this?
Question 38A team labels a dataset where each record includes both the input features and the known correct output, then trains a model to predict that output. Which type of machine learning is this?
Question 39A company needs an application to read generated text aloud in a natural-sounding voice for an accessibility feature. Which AWS service converts text to lifelike speech?
Question 40A company wants to customize a foundation model on Amazon Bedrock so it adapts to their proprietary domain by adjusting the model's weights using a labeled dataset of examples. Which approach is this?
Question 41A healthcare provider must keep all data used with AI services encrypted, and wants to manage and rotate its own encryption keys. Which AWS service should be used to manage these keys?
Question 42A data governance team wants to automatically discover and classify sensitive data such as PII stored in Amazon S3 buckets that may feed AI training pipelines. Which AWS service is designed for this?
Question 43A team wants a model that answers questions using the company's private documents without retraining the model, by retrieving relevant document passages at query time and adding them to the prompt. Which technique should they use?
Question 44A model that approves or denies loan applications is found to systematically reject qualified applicants from a particular demographic group. Which Responsible AI concern does this MOST directly represent, and which AWS tool helps detect it?
Question 45A company wants to reduce hallucinations in a customer-support chatbot built on a Bedrock foundation model by grounding answers in its internal product documentation, without retraining the model. Which approach best meets this requirement?
Question 46A team needs to adapt a foundation model so it consistently answers in their company's specialized legal terminology and tone. They have several thousand high-quality prompt-and-response pairs. Which technique is most appropriate?
Question 47Which characteristic best distinguishes a foundation model from a traditional task-specific machine learning model?
Question 48An AI engineer is building an agent on Amazon Bedrock that must look up order status by calling an internal REST API during a conversation. Which Bedrock capability enables the model to invoke this external API as part of completing the task?
Question 49A company deploys a generative AI summarization feature and is concerned about unpredictable monthly Bedrock costs as usage grows. Which factor most directly drives the cost of on-demand Bedrock inference?
Question 50A RAG application returns answers that are factually correct but often irrelevant to the user's actual question. The retrieval step is suspected. Which improvement most directly targets retrieval relevance?
Question 51A team wants to evaluate the quality of summaries generated by different foundation models at scale, without hiring large numbers of human reviewers for every comparison. Which evaluation approach is commonly used?
Question 52A startup needs occasional, low-volume access to a large foundation model and wants to avoid paying for idle reserved capacity. Which Bedrock pricing model best fits this usage pattern?
Question 53An application has steady, high-volume, latency-sensitive traffic to a single Bedrock model and needs guaranteed throughput. Which option is most appropriate?
Question 54A team uses Amazon Bedrock Knowledge Bases to power a RAG chatbot. What is the primary role of the Knowledge Base in this architecture?
Question 55During inference, raising the temperature parameter of a generative model from 0.2 to 0.9 will most likely:
Question 56A company wants to monitor its production Bedrock workload for usage metrics, latency, and errors, and to keep an auditable log of API calls. Which combination of AWS services best supports this?
Question 57A developer wants to improve a foundation model's answers on a niche task by including a few worked examples directly in the prompt, without any training. This technique is known as:
Question 58A bank wants to add guardrails to a Bedrock-powered assistant to block harmful content and prevent it from revealing personally identifiable information (PII). Which AWS capability is designed for this?
Question 59A team is comparing fine-tuning versus RAG for keeping a chatbot's answers current as the company's policy documents change weekly. Which statement is most accurate?
Question 60What is the primary purpose of converting text into embeddings (vector representations) in a RAG pipeline?
Question 61A company must choose between two foundation models for a high-volume task. Model A is more accurate but 5x more expensive per token; Model B is slightly less accurate but much cheaper. The task tolerates minor accuracy differences. What is the most cost-effective recommendation?
Question 62An ML team observes that a deployed model's prediction quality is slowly degrading over time as real-world input data shifts away from the training distribution. What is this phenomenon called, and what monitoring practice addresses it?
Question 63A retailer wants to deploy a generative AI feature but must ensure that the prompts and data it sends are not used to train the base foundation models and remain within its control. Which consideration is most relevant when choosing Amazon Bedrock?
Question 64To reduce both latency and cost for repeated, identical generative AI requests (for example, the same system prompt processed many times), which technique is most effective?
Question 65An evaluation report for a generative model shows high scores on automated metrics like ROUGE and BLEU, but business users complain the outputs feel unhelpful and off-tone. What is the best next step?