AIF-C01 Practice Test 37 — 15 Questions

Question 1

A company is building a customer support chatbot using Amazon Bedrock. They need to store conversation history for context across sessions. Which AWS service is best suited for this purpose?

Accepted Answer

Amazon DynamoDB. Amazon DynamoDB is the best choice for storing conversation history because it is a fully managed NoSQL key-value and document database that provides single-digit millisecond latency at any scale. It supports flexible schema, which is ideal for storing variable-length chat sessions, and its Time to Live (TTL) feature can automatically expire old conversations to manage storage costs. DynamoDB also integrates natively with AWS Lambda and Amazon Bedrock for real-time retrieval and update of context across sessions.

Answer

Amazon S3

Answer

Amazon RDS

Answer

Amazon ElastiCache

Question 2

What is the primary purpose of an embedding model in the context of RAG?

Accepted Answer

Convert text into dense vector representations for semantic search. In Retrieval-Augmented Generation (RAG), the embedding model's primary role is to convert text (e.g., documents or queries) into dense vector representations. These vectors capture semantic meaning, enabling efficient similarity search in vector databases to retrieve the most relevant context for the LLM. Option B correctly identifies this core function.

Answer

Generate human-readable summaries of documents

Answer

Rank retrieved documents before passing them to the LLM

Answer

Generate final answers from retrieved context

Question 3

A developer is new to Amazon Bedrock and wants to understand the components of tokenization and context windows. Which TWO statements are correct? (Select TWO.)

Accepted Answer

The context window determines the maximum number of tokens the model can process in a single request. Option C is correct because the context window defines the hard limit on the total number of tokens (input plus output) that a foundation model can handle in one inference request on Amazon Bedrock, so exceeding it requires truncation or chunking. Option E is correct because modern tokenizers such as BPE or SentencePiece break text into tokens that may be whole words, subwords, punctuation, or characters, not fixed units. Option A is wrong because a larger context window only increases how much text can be supplied; response quality still depends on prompt design, model capability, and relevance of the included content. Option B is wrong because tokenization does not split solely on whitespace; subword tokenizers split within words and handle punctuation and special characters. Option D is wrong because Bedrock pricing is based on the number of input and output tokens processed (per 1,000 tokens), not on tokens per second, which is a throughput metric.

Answer

A larger context window always produces better quality responses

Answer

Tokenization splits text only on whitespace

Answer

The cost of a request is based on the number of tokens per second

Question 4

A company is using Amazon Bedrock to run batch inference on millions of customer support transcripts for sentiment analysis. Which approach is MOST cost‑effective and fastest for this workload?

Accepted Answer

Use Bedrock batch inference to process the transcripts asynchronously. Bedrock batch inference is designed for large-scale, asynchronous processing of millions of records, making it the most cost-effective and fastest approach for this workload. It processes data in bulk without requiring real-time latency, and you pay only for the compute time used, avoiding the per-token costs of real-time APIs. This eliminates the need to manage throughput or caching for a non-interactive batch job.

Answer

Use the Bedrock real‑time API with on‑demand throughput

Answer

Enable model caching on the real‑time API to reuse previous responses

Answer

Use the Bedrock real‑time inference API with a provisioned throughput model

Question 5

A developer is using Amazon Bedrock to generate text responses. They want to reduce the randomness of the output and make the model more deterministic. Which parameter should the developer decrease?

Accepted Answer

Temperature. Temperature directly controls the randomness of the model's output. Lowering the temperature (e.g., from 1.0 to 0.2) reduces the probability of sampling less likely tokens, making the model more deterministic and focused on the highest-probability next token. This is the standard parameter for adjusting output creativity versus determinism in large language models like those on Amazon Bedrock.

Answer

stop_sequences

Answer

max_tokens

Answer

top_p

Question 6

An AI practitioner is evaluating foundation models on Amazon Bedrock for a text summarization task. The input documents average 6,000 tokens. The model must process the entire document in a single pass without chunking. Which model capability is MOST critical for this requirement?

Accepted Answer

The context window size. The context window size determines the maximum number of tokens a model can process in a single forward pass. Since the input documents average 6,000 tokens and must be processed without chunking, the model's context window must be at least 6,000 tokens. Amazon Bedrock models like Claude 3 have context windows of 100K or 200K tokens, making this capability critical for the requirement.

Answer

The output modality (text vs. multimodal)

Answer

The temperature setting

Answer

The number of parameters in the model

Question 7

A company is evaluating the performance of a summarization model using Amazon Bedrock model evaluation. They want an automated metric that measures how well the generated summary captures the meaning of the reference summary. Which metric is MOST suitable?

Accepted Answer

BERTScore. BERTScore is the most suitable metric because it uses contextual embeddings from BERT to compute semantic similarity between generated and reference summaries, capturing meaning rather than exact n-gram overlap. This makes it ideal for summarization evaluation where paraphrasing and semantic equivalence are critical.

Answer

BLEU

Answer

ROUGE-L

Answer

Accuracy

Question 8

A company is building a generative AI application using Amazon Bedrock that requires low-latency responses for a global user base. They need to select a vector store for the knowledge base. Which two options are fully managed AWS services suitable for this requirement? (Choose TWO.)

Accepted Answer

Amazon Aurora with pgvector extension. Amazon OpenSearch Serverless (E) is a fully managed AWS service with a vector search collection type designed for high-scale, low-latency similarity queries, and it is natively supported as a Bedrock knowledge base vector store. Amazon Aurora with pgvector extension (B) is a fully managed AWS relational database service that supports vector similarity search via the pgvector extension, but it is not optimized for global low-latency vector search at scale. Pinecone (A) and MongoDB Atlas (C) are third-party services, not fully managed AWS services. Amazon DynamoDB (D) is a fully managed AWS service but does not natively provide vector similarity search for Bedrock knowledge bases. Therefore, the two fully managed AWS services suitable for this requirement are Amazon OpenSearch Serverless and Amazon Aurora with pgvector, but the low-latency global requirement makes OpenSearch Serverless the primary choice. However, since the question asks for two options, both B and E are the only AWS-native vector stores listed, so they remain the correct choices despite the latency nuance.

Answer

Pinecone

Answer

MongoDB Atlas

Answer

Amazon DynamoDB

Question 9

A developer is building a RAG application using Amazon Aurora PostgreSQL with the pgvector extension. After ingesting documents, the vector search returns results that are not always the most relevant. What should the developer adjust to improve relevance?

Accepted Answer

Switch the vector similarity metric to cosine similarity if using Euclidean, or adjust chunk size. Relevance in vector search is primarily determined by the similarity metric and chunking strategy. Cosine similarity is often more effective than Euclidean for semantic similarity in high-dimensional spaces, and adjusting chunk size ensures that each vector captures a coherent semantic unit, directly improving retrieval relevance.

Answer

Enable pgvector index type to IVFFlat with a low number of lists

Answer

Change the PostgreSQL instance to a larger instance type

Answer

Increase the number of provisioned Aurora read replicas

Question 10

A company is using Amazon Bedrock to build a sentiment analysis application for customer reviews. They need to evaluate the model's performance against a labeled test dataset. They want to use a metric that compares the model's predicted sentiment (positive, negative, neutral) to the ground truth labels. Which metric is MOST appropriate?

Accepted Answer

Accuracy. Accuracy is the most appropriate metric for this multi-class classification task because it directly measures the proportion of correctly predicted sentiments (positive, negative, neutral) out of total predictions, given a labeled test dataset. It is simple, interpretable, and well-suited for evaluating categorical outputs where class distribution is balanced or the cost of misclassification is uniform.

Answer

ROUGE-1

Answer

BERTScore

Answer

BLEU

Question 11

An enterprise is building a RAG solution with Amazon Bedrock and needs to ensure that the retrieved documents are from authorised sources only. They also must prevent the model from generating responses that contain personally identifiable information (PII). Which two Bedrock features combined address these requirements?

Accepted Answer

Amazon Bedrock Knowledge Bases and Bedrock Guardrails. Amazon Bedrock Knowledge Bases lets you ground responses in your own curated data sources (e.g., S3, OpenSearch), so retrieval is limited to the documents you configure and index. Amazon Bedrock Guardrails provides configurable safeguards, including sensitive information filters that can detect and block PII in model responses. Combining Knowledge Bases for controlled retrieval with Guardrails for PII filtering addresses both requirements.

Answer

Amazon Bedrock Model Evaluation and Bedrock Studio

Answer

Amazon Bedrock Agents and Bedrock Guardrails

Answer

Amazon Bedrock Playground and prompt engineering

Question 12

An AI team is evaluating a text generation model for a clinical summarization task. They have a set of 100 sample summaries and corresponding expert-written reference summaries. The team wants both automated metrics and human evaluation. Which THREE metrics are appropriate for evaluating summarization quality? (Choose THREE)

Accepted Answer

ROUGE. ROUGE (A) is correct because it is the standard automated metric for summarization, measuring n-gram overlap (ROUGE-N), longest common subsequence (ROUGE-L), and skip-bigram (ROUGE-S) recall between generated and reference summaries. Human evaluation by domain experts (B) is correct because clinical summarization requires judging factual accuracy, clinical relevance, and fluency, which automated n-gram metrics cannot reliably capture, especially for safety-critical medical content. BERTScore (D) is correct because it uses contextual embeddings to compute semantic similarity between generated and reference summaries, capturing meaning-level matches that ROUGE's surface overlap misses. Perplexity (C) is not appropriate here because it measures a language model's likelihood of a token sequence and does not compare generated text against reference summaries, so it does not assess summarization quality. BLEU (E) is not appropriate because it was designed for machine translation and emphasizes precision of n-grams, correlating poorly with summarization quality compared to recall-oriented ROUGE.

Answer

Perplexity

Answer

BLEU

Question 13

A corporation needs to deploy a chatbot that must not output any personally identifiable information (PII) such as credit card numbers or social security numbers. Which TWO Bedrock features should they use together to achieve this? (Select TWO.)

Accepted Answer

Bedrock Guardrails with PII detection. Amazon Bedrock Guardrails provides PII detection and redaction that identifies sensitive data patterns such as credit card numbers and social security numbers in prompts and model responses, allowing the chatbot to block or mask that information. Option E is not a valid Bedrock feature for PII redaction; using a Bedrock Agent with a Lambda function to post-process responses is a custom integration pattern, not a standard Bedrock capability for preventing PII output. The options as written do not provide a valid second Bedrock feature, so only D is marked correct.

Answer

Bedrock Knowledge Bases with grounding

Answer

Bedrock Guardrails with content filtering

Answer

Bedrock Guardrails with topic denial

Answer

Bedrock Agent with a Lambda function to post-process responses

Question 14

A financial services company is deploying a machine learning model to approve loans. They want to ensure that the model does not discriminate based on race or gender. Which AWS service or feature can help them detect bias in the model's predictions?

Accepted Answer

Amazon SageMaker Clarify. Amazon SageMaker Clarify is purpose-built to detect bias in ML models and datasets. It computes bias metrics such as Disparate Impact (DI) and Difference in Conditional Acceptance (DCA) across protected attributes like race and gender, and also provides explainability via SHAP values. This directly addresses the requirement to detect discrimination in loan approval predictions.

Answer

AWS CloudTrail

Answer

Amazon SageMaker Model Monitor

Answer

AWS Identity and Access Management (IAM)

Question 15

A company is deploying a large language model (LLM) for customer support. They want to reduce the risk of hallucinations. Which TWO approaches should they implement? (Choose two.)

Accepted Answer

Implement Amazon Bedrock Guardrails to define topics the model should avoid. Option A is correct because Amazon Bedrock Guardrails lets you define denied topics, content filters, and contextual grounding checks that constrain what the model can say, directly reducing the chance it produces off-topic or fabricated responses in a customer support setting. Option E is correct because Retrieval-Augmented Generation (RAG) retrieves authoritative documents from a knowledge base and injects that factual context into the prompt, grounding the model's answers in real data rather than relying solely on parametric memory, which is the standard mitigation for hallucinations. Option B is wrong because a larger model without retrieval still generates from learned parameters and can hallucinate confidently; scale alone does not guarantee factual accuracy. Option C is wrong because fine-tuning on hallucinated examples would teach the model to reproduce fabrications, worsening the problem. Option D is wrong because increasing max token length only allows longer outputs and does not improve factual grounding; it can even give the model more room to elaborate on incorrect claims.

Answer

Use a larger model without any retrieval mechanism

Answer

Fine-tune the model on a dataset of hallucinated examples

Answer

Increase the maximum token length to give the model more room to elaborate