CertifyDeckCargando contenido…
AIF-C01 · Sample questions
10 real questions from the CertifyDeck AIF-C01 bank, spread across the domains of the official AWS Certified AI Practitioner skills outline, with the correct answer and a full explanation of why the other options are wrong. They are original questions written from the public syllabus, not exam dumps, and they are the same style you will find in the practice tests and mock exams.
Correct answer: A. Stratified sampling
Stratified sampling divides a population into subgroups, or strata, based on shared characteristics and takes a random sample from each one, which helps a dataset represent every group. K-fold cross-validation is tempting because it also splits data, but it splits training data into folds to estimate how well a model generalizes, not to balance demographic groups. Early stopping is a training technique against overfitting, and deduplication removes repeated records.
Correct answer: C. Less accurate predictions for younger and older customers
Bias can come from the data as well as from the algorithm: a model trained primarily on data from middle-aged individuals may be less accurate when it makes predictions for younger and older people. The effect is not uniform across groups, because the well-represented group is learned better. Fewer examples do not make a group's predictions more accurate, and the claim that data cannot introduce bias contradicts the documentation.
Correct answer: A. Conversational capabilities
Generative AI can respond naturally to human conversation, which is what lets chatbots and virtual assistants engage customers and resolve requests on first contact. Deterministic output is not an advantage of generative AI: models sample from a probability distribution, so the same prompt can produce different responses. Interpretability is a documented limitation, because these models are often considered black boxes.
Correct answer: D. Track ROI continuously on a dashboard that combines costs and business value
AWS Prescriptive Guidance says that ROI for a generative AI application must be managed as a dynamic KPI on a dashboard that combines financial metrics, such as cost per interaction and infrastructure spend, with business value metrics, such as hours saved, revenue lift and CSAT. A one-time calculation at launch misses that both costs and value change over time. Tracking only spend or only accuracy shows one side of the equation, not the return.
Correct answer: A. AWS Glue Data Quality
AWS Glue Data Quality evaluates rules written in DQDL to produce a data quality score, and in AWS Glue ETL jobs it can identify and filter out bad data before it is loaded into the data lake. Lake Formation is tempting because it governs the data lake, but it manages access permissions, not data quality. SageMaker Model Monitor checks data captured from deployed model endpoints rather than data being loaded into a data lake, and S3 Object Lock prevents objects from being changed or deleted.
Correct answer: B. Enable Amazon Bedrock model invocation logging
Model invocation logging is disabled by default; once enabled, it collects the full request data, response data and metadata of invocations and publishes them to Amazon CloudWatch Logs, Amazon S3 or both. CloudTrail is tempting because it records Bedrock API calls, but its entries show who called which operation and when, not the full prompt and response text. GuardDuty detects threats, and Inspector scans workloads for software vulnerabilities.
Correct answer: B. Amazon SageMaker AI managed model endpoints
SageMaker AI managed model endpoints provide more flexibility than Bedrock's fully managed hosting and less operational overhead than a completely self-managed solution on EC2. Bedrock on-demand invocation is the fully managed option that offers the least control over hosting. Self-managed EC2 instances give full control but require the team to run its own servers, and Knowledge Bases supplies retrieved context rather than hosting a model.
Correct answer: C. Scalability and 24/7 availability
AI-powered chatbots and virtual assistants can automate customer support so that a solution scales with demand and answers around the clock without the limits of human staffing. Explainability, fairness, and veracity are responsible AI dimensions that any AI solution should address, but none of them is the value that lets the retailer absorb peaks in volume at any hour.
Correct answer: C. Enable the pgvector extension in the existing database
pgvector is an open-source PostgreSQL extension that adds storage and search of vector embeddings, so Amazon RDS for PostgreSQL can run relational SQL queries and vector similarity searches in the same database. Neptune Analytics, MemoryDB and S3 Vectors all support vectors, but each is a separate store, which fails the requirement to keep the embeddings in the existing database.
Correct answer: A. Amazon Neptune Analytics
Amazon Neptune Analytics is a graph analytics engine with native vector search, which suits GraphRAG use cases that traverse complex relationships alongside vector similarity and give explainable answers. Amazon RDS for PostgreSQL with pgvector fits combined relational and vector queries, and Amazon S3 Vectors fits low-cost storage with infrequent queries; neither offers graph traversal.
Answer each question before opening the explanation, and write down why you discarded the other options. On AIF-C01, most mistakes come from choosing an option that is technically possible but does not meet every requirement in the scenario.
These 10 questions are a taste of the format, not a diagnostic. To know which domains you fail, practice by domain for free and take a timed mock exam that mirrors the official weights and duration.
Start for free, no sign-up or card. Every question includes an explanation like the ones above, and the mock exam reproduces the official AIF-C01 format.