Pass the AWS AI Practitioner — the First Time

1. A data science team describes deep learning to a new stakeholder. Which statement best characterizes how deep learning relates to artificial intelligence (AI) and machine learning (ML)?

   • A. Deep learning is a broad field that contains both AI and ML as subfields.
   • B. Deep learning is a subset of ML, which is itself a subset of the broader field of AI.
   • C. Deep learning and ML are unrelated disciplines that both fall outside of AI.
   • D. ML is a subset of deep learning, which is a subset of AI.

   Answer: B. AI is the broadest field, ML is a subset of AI that learns patterns from data, and deep learning is a subset of ML that uses multi-layered neural networks. Option A and D invert the hierarchy, and C wrongly excludes them from AI.

   Exam domain: Fundamentals of AI and ML

2. An online retailer has a dataset of customer transactions where each record includes a label indicating whether the transaction was fraudulent or legitimate. The team wants a model to predict fraud on new transactions. Which type of machine learning does this represent?

   • A. Unsupervised learning
   • B. Reinforcement learning
   • C. Self-organizing learning
   • D. Supervised learning

   Answer: D. Supervised learning trains on labeled data (each example has a known target, here fraud or legitimate) to predict labels for new data. Unsupervised learning works on unlabeled data to find structure, reinforcement learning learns from rewards through trial-and-error interaction, and 'self-organizing learning' is not a standard ML category (self-organizing maps are actually an unsupervised technique), so none match a labeled fraud dataset.

   Exam domain: Fundamentals of AI and ML

3. A logistics company collects sensor readings every minute from delivery trucks, producing values such as temperature and speed indexed by timestamp. To forecast next week's fuel consumption, an analyst needs to choose the correct way to describe this data. Which data type best describes these readings?

   • A. Unstructured image data
   • B. Categorical text data
   • C. Geospatial vector data
   • D. Time-series data

   Answer: D. Time-series data consists of observations recorded sequentially over time at consistent intervals, exactly matching per-minute timestamped sensor values used for forecasting. Image data is unstructured pixels, text data is unstructured language, and while location may appear, the defining characteristic here is the time-indexed sequence.

   Exam domain: Fundamentals of AI and ML

4. A team has trained a model and deployed it. Once per night, they collect all of the day's accumulated records and run them through the model in a single large job to generate predictions, which are written to a database before morning. Which inferencing approach are they using?

   • A. Streaming inferencing
   • B. Online inferencing
   • C. Batch inferencing
   • D. Real-time inferencing

   Answer: C. Batch inferencing processes a large group of accumulated records at scheduled intervals rather than responding to individual requests instantly, which fits a nightly bulk job. Real-time, streaming, and online inferencing all imply low-latency, per-request predictions as data arrives, which is not what is described.

   Exam domain: Fundamentals of AI and ML

5. A developer is integrating a large language model (LLM) into an application and learns that the model processes text by first breaking it into smaller units that are mapped to numeric IDs before any computation occurs. What are these smaller units called?

   • A. Parameters
   • B. Tokens
   • C. Embeddings
   • D. Vectors

   Answer: B. Tokens are the discrete units (words, sub-words, or characters) that an LLM splits text into and maps to numeric IDs before processing. Embeddings are dense numeric representations that capture semantic meaning, and vectors are the mathematical arrays embeddings are stored as, but those come after tokenization. Parameters are the learned weights of the model, not units of input text.

   Exam domain: Fundamentals of Generative AI

6. A media company wants to generate photorealistic marketing images from natural-language text descriptions. Which type of generative AI model is specifically designed for this kind of high-quality image synthesis?

   • A. A decision tree ensemble
   • B. A k-means clustering model
   • C. A diffusion model
   • D. A linear regression model

   Answer: C. Diffusion models generate images by progressively denoising random noise into a coherent image guided by a text prompt, and are the dominant architecture for text-to-image generation (for example, Stable Diffusion). Decision tree ensembles, linear regression, and k-means are traditional supervised or unsupervised ML methods that do not perform generative image synthesis.

   Exam domain: Fundamentals of Generative AI

7. A team is building a retrieval-augmented generation (RAG) solution. Before storing a large library of PDF policy documents in a vector database, they must first split each document into smaller, semantically coherent passages so that retrieval returns relevant context. What is this preprocessing step called?

   • A. Chunking
   • B. Fine-tuning
   • C. Quantization
   • D. Tokenization

   Answer: A. Chunking is the process of dividing large documents into smaller passages so each piece can be embedded and retrieved individually in a RAG pipeline. Tokenization splits text into model input tokens (a different granularity), fine-tuning updates model weights, and quantization reduces numeric precision to shrink a model; none describe splitting documents into retrievable passages.

   Exam domain: Fundamentals of Generative AI

8. A logistics company has about 40 internal REST APIs described by OpenAPI specifications plus a dozen AWS Lambda functions. Agents built with several frameworks must discover and call these as MCP tools through one managed endpoint, with inbound and outbound authentication handled for them. The company does not want to write and host a separate MCP server for each API. Which service meets these requirements?

   • A. Amazon Bedrock AgentCore Runtime
   • B. Amazon Bedrock AgentCore Gateway
   • C. Amazon Bedrock AgentCore Memory
   • D. Amazon Bedrock AgentCore Identity

   Answer: B. AgentCore Gateway converts APIs (OpenAPI or Smithy) and Lambda functions into MCP-compatible tools behind a single secure endpoint, with built-in tool discovery and both ingress and egress authentication. Runtime hosts and scales the agent code itself. Memory stores conversation and long-term context. Identity manages agent and user credentials, but it does not turn APIs into tools.

   Exam domain: Fundamentals of Generative AI

9. A startup wants to add a chatbot to its customer support portal. Support tickets often reference the company's internal product manuals, which change weekly. The team wants the chatbot to answer using the most current manual content without retraining a model each week. Which approach on Amazon Bedrock best meets this requirement?

   • A. Increase the model's temperature so it generates more creative answers
   • B. Continuously pre-train a foundation model from scratch on the manuals
   • C. Fine-tune a foundation model on the manuals every week
   • D. Use RAG with a Bedrock Knowledge Base over the manuals

   Answer: D. RAG retrieves up-to-date content from an external knowledge source at inference time, so weekly manual updates are reflected immediately by re-indexing rather than retraining. Bedrock Knowledge Bases manage the ingestion, chunking, embedding, and retrieval pipeline. Fine-tuning and continuous pre-training are costly, slow, and would need to be repeated for every change; temperature only affects output randomness, not factual freshness.

   Exam domain: Applications of Foundation Models

10. An e-commerce company is building a RAG application on Amazon Bedrock. The team has already chosen a relational database for its transactional data and wants to store and query document embeddings in the SAME managed relational engine to minimize the number of services to operate. Which AWS option natively supports vector storage and similarity search for this requirement?

    • A. Amazon Redshift with a sort key on the embedding column
    • B. Amazon Aurora PostgreSQL with the pgvector extension
    • C. Amazon ElastiCache for Memcached caching
    • D. Amazon DynamoDB with a global secondary index

    Answer: B. Aurora PostgreSQL-Compatible Edition supports the pgvector extension, allowing vector embeddings to be stored and queried with similarity search inside a managed relational engine, satisfying the desire to reuse the relational platform. It is also a supported vector store for Amazon Bedrock Knowledge Bases. DynamoDB is a NoSQL key-value/document store without native vector similarity search, Redshift is an analytics warehouse not designed as a low-latency vector store for RAG, and ElastiCache for Memcached is an in-memory cache without vector search.

    Exam domain: Applications of Foundation Models

Want more? Finish the quiz above to unlock 10 more free questions, or open the full AIF-C01 practice app.