AWS Generative AI Develop - Free Practice Quiz

Question 1 of 60

Which Amazon Bedrock feature provides a fully managed RAG pipeline that automatically handles chunking, embedding, and vector storage from S3 data sources?

A Bedrock Knowledge Bases B Bedrock Agents C Bedrock Model Evaluation D Bedrock Fine-tuning

1 / 60

Question 2 of 60

A developer is building a RAG application and notices that retrieved chunks often contain only partial answers because the chunks are too small. What chunking strategy adjustment should they make?

A Use a different embedding model B Switch to character-based chunking C Reduce the overlap between chunks D Increase the chunk size or use a semantic chunking strategy

2 / 60

Question 3 of 60

Which inference parameter controls the randomness of a foundation model's output, where a lower value produces more deterministic responses?

A Temperature B Top-k C Max tokens D Top-p

3 / 60

Question 4 of 60

A company wants to build an agent that can look up real-time stock prices and then execute trades based on the result. Which Amazon Bedrock feature enables the agent to call external APIs as part of its reasoning loop?

A Bedrock Model Evaluation B Bedrock Guardrails C Bedrock Knowledge Bases D Bedrock Agents with Action Groups

4 / 60

Question 5 of 60

What is the ReAct pattern in the context of agentic AI?

A A React.js framework for building AI front-ends B A pattern where an agent interleaves reasoning steps and actions in a loop until a task is complete C A pattern for chaining multiple LLM calls sequentially D A method for fine-tuning models with reinforcement learning

5 / 60

Question 6 of 60

Which AWS vector store is natively integrated with Amazon Bedrock Knowledge Bases and provides serverless vector search at scale?

A Amazon RDS for PostgreSQL B Amazon ElastiCache C Amazon DynamoDB D Amazon OpenSearch Serverless

6 / 60

Question 7 of 60

A developer wants to use pgvector for storing embeddings in a RAG application. Which AWS database service supports this?

A Amazon Redshift B Amazon Aurora PostgreSQL C Amazon DynamoDB D Amazon Neptune

7 / 60

Question 8 of 60

What is the primary trade-off between using RAG versus fine-tuning a foundation model to incorporate company-specific knowledge?

A Fine-tuning cannot be done on AWS B RAG requires retraining the model whenever new data is added C RAG provides dynamic, updatable knowledge without modifying model weights, while fine-tuning embeds static knowledge into the model D RAG is more expensive and slower to implement

8 / 60

Question 9 of 60

Which parameter limits the total number of tokens the model can generate in a single response?

A Top-p B Top-k C Max tokens D Temperature

9 / 60

Question 10 of 60

A developer is experiencing high latency in their Bedrock application during peak traffic. They want to guarantee a minimum level of throughput. Which Bedrock feature should they configure?

A Model fine-tuning B Provisioned Throughput C Bedrock Guardrails D Bedrock Knowledge Bases

10 / 60

Question 11 of 60

In a RAG pipeline, what is the purpose of the 'augmentation' step?

A Injecting the retrieved context into the prompt sent to the LLM B Selecting the most relevant chunks from the vector store C Storing the LLM's response in a database D Converting documents into vector embeddings

11 / 60

Question 12 of 60

Which Amazon Bedrock Guardrails capability can automatically detect and redact sensitive information like credit card numbers and social security numbers from model inputs and outputs?

A Grounding checks B Denied topics C Content filters D PII redaction

12 / 60

Question 13 of 60

A generative AI application is producing responses that are not grounded in the retrieved documents and contains fabricated information. Which Bedrock Guardrails feature can help detect and block these ungrounded responses?

A Grounding checks B Content filters C PII redaction D Denied topics

13 / 60

Question 14 of 60

What is the primary purpose of the 'system prompt' in a foundation model interaction?

A To define the embedding model to use B To provide few-shot examples to the model C To set persistent behavioral instructions, persona, and constraints for the model D To specify the maximum number of tokens to generate

14 / 60

Question 15 of 60

A developer is selecting a foundation model for a document summarization task. The documents are 50,000 tokens long. Which model characteristic is most critical to evaluate first?

A Context window size B Model parameter count C Model fine-tuning support D Training data cutoff date

15 / 60

Question 16 of 60

Which metric is commonly used to evaluate the quality of text generated by AI models by comparing n-gram overlaps with reference texts?

A Perplexity B F1 Score C ROUGE D RAGAS

16 / 60

Question 17 of 60

A team is building a customer support agent that needs to retrieve answers from a product manual and also check order status via an API. What is the correct Amazon Bedrock architecture?

A Bedrock Fine-tuning with the product manual as training data B Bedrock Knowledge Bases only C Two separate Bedrock Knowledge Bases D Bedrock Agents with a Knowledge Base for the manual and an Action Group for the order status API

17 / 60

Question 18 of 60

What does the 'top-p' (nucleus sampling) inference parameter control?

A The number of candidate tokens considered at each step B The penalty applied to repeated tokens C The maximum response length D The cumulative probability threshold of tokens considered for sampling

18 / 60

Question 19 of 60

A developer wants to enable streaming responses in their Bedrock application so users see text as it is generated rather than waiting for the full response. Which API should they use?

A InvokeModelWithResponseStream B InvokeModel C GetModel D StartModelInvocationJob

19 / 60

Question 20 of 60

Which AWS service should be configured to capture all Amazon Bedrock model invocation inputs and outputs for auditing and debugging purposes?

A AWS Config B Bedrock Model Invocation Logging C Amazon CloudWatch Metrics D AWS CloudTrail

20 / 60

Question 21 of 60

In a RAG system, what problem does 'chunk overlap' solve?

A It reduces vector storage costs B It ensures that context spanning chunk boundaries is not lost during retrieval C It reduces the total number of chunks needed D It improves embedding model performance

21 / 60

Question 22 of 60

A developer needs to fine-tune a Llama model on Amazon Bedrock with their company's proprietary dataset to improve performance on a specialized legal reasoning task. What data format is typically required for instruction fine-tuning?

A CSV with labeled categories B Prompt-completion pairs in JSONL format C Raw text paragraphs D SQL table format

22 / 60

Question 23 of 60

Which RAGAS metric evaluates whether the answer generated by an LLM is actually supported by the retrieved context?

A Faithfulness B Context Precision C Context Recall D Answer Relevancy

23 / 60

Question 24 of 60

An agent is performing a multi-step task and needs to remember information from earlier steps to use in later steps within the same session. Which Amazon Bedrock Agents feature supports this?

A Knowledge Base retrieval B Action Groups C Session attributes and memory D Guardrails

24 / 60

Question 25 of 60

What is 'prompt chaining' in the context of generative AI application design?

A A technique for reducing prompt token count B Storing prompts in a chain database C Breaking a complex task into a sequence of LLM calls where each output feeds the next prompt D Combining multiple prompts into a single API call

25 / 60

Question 26 of 60

A company wants to prevent their Bedrock-powered customer service bot from responding to queries about competitor products. Which guardrail configuration should they implement?

A PII redaction for competitor names B Grounding checks against a knowledge base C A denied topic defining competitor-related discussions D Content filters set to HIGH

26 / 60

Question 27 of 60

Which approach is most cost-effective for a use case where the same long system prompt must be reused across thousands of API calls?

A Fine-tuning the model to embed the system prompt B Using Bedrock Prompt Caching to cache the system prompt C Sending the full system prompt with every API call D Storing the system prompt in a Knowledge Base

27 / 60

Question 28 of 60

A developer is comparing two retrieval strategies for their RAG system: dense retrieval using embeddings versus sparse retrieval using keyword matching (BM25). A hybrid approach combines both. What is the primary advantage of hybrid retrieval?

A Simpler implementation than either approach B Better handling of both semantic similarity queries and exact keyword matches C Lower cost than either approach alone D Faster retrieval than dense retrieval alone

28 / 60

Question 29 of 60

In the context of Bedrock Knowledge Bases, what is the role of the embedding model?

A To convert document chunks into vector representations for storage and similarity search B To rank retrieved chunks by relevance C To generate the final answer from retrieved chunks D To split documents into appropriately sized chunks

29 / 60

Question 30 of 60

A developer notices their Bedrock agent sometimes takes too many reasoning steps and exceeds the maximum orchestration steps limit. What is the most appropriate architectural fix?

A Disable the agent's memory B Use a smaller foundation model C Increase the temperature parameter D Break the task into sub-agents with a supervisor agent

30 / 60

Question 31 of 60

Which foundation model characteristic best determines its suitability for tasks requiring deep reasoning and complex instruction following compared to simple text generation?

A Maximum tokens per minute throughput B Image input support C Model size (parameter count) and training methodology D Number of languages supported

31 / 60

Question 32 of 60

What is the key architectural difference between Bedrock Agents and a simple prompt chain?

A Bedrock Agents cannot call external APIs B Bedrock Agents use a different API endpoint C Prompt chains are always faster than agents D Bedrock Agents autonomously decide which tools to use and in what order based on the task; prompt chains use a predetermined sequence

32 / 60

Question 33 of 60

A team wants to evaluate their RAG system's retrieval quality specifically. Which RAGAS metric measures whether the retrieved context contains all the information needed to answer the question?

A Faithfulness B Context Precision C Context Recall D Answer Relevancy

33 / 60

Question 34 of 60

Which parameter would a developer adjust to make a foundation model's responses more creative and varied for a creative writing application?

A Increase max tokens B Increase temperature C Decrease temperature D Decrease top-p

34 / 60

Question 35 of 60

A security team requires that all prompts sent to Amazon Bedrock and all model responses are logged for compliance purposes. Where should Model Invocation Logging data be sent for long-term retention and analysis?

A Amazon S3 and/or Amazon CloudWatch Logs B AWS Secrets Manager C Amazon DynamoDB D AWS CloudTrail only

35 / 60

Question 36 of 60

When building a multi-turn conversational application with a foundation model, what must the developer include in each API request to maintain conversation context?

A The system prompt only B The full conversation history (all prior turns) in the messages array C Only the latest user message D A session ID parameter

36 / 60

Question 37 of 60

A developer is choosing between on-demand pricing and Provisioned Throughput for their Bedrock application. When is Provisioned Throughput the better choice?

A For applications with consistent high-volume traffic where predictable latency and throughput guarantees are required B When using models available for free tier C For applications with low, sporadic traffic D For development and testing environments

37 / 60

Question 38 of 60

Which AWS IAM best practice should be applied when a Lambda function needs to invoke Amazon Bedrock models on behalf of a Bedrock Agent's Action Group?

A Assign a least-privilege IAM execution role to the Lambda function B Use long-term access keys stored in the Lambda environment variables C Grant the Lambda function full AdministratorAccess D Use the root account credentials

38 / 60

Question 39 of 60

A developer wants to build a RAG application where the knowledge base is updated in real time as new documents arrive in S3. What is the correct Bedrock Knowledge Bases configuration?

A Use Amazon Kendra instead, as Bedrock Knowledge Bases do not support real-time updates B Configure an S3 event notification to trigger a Knowledge Base sync via the StartIngestionJob API C Manually re-sync the knowledge base weekly D Use Bedrock Fine-tuning instead of Knowledge Bases for real-time data

39 / 60

Question 40 of 60

What is 'function calling' (tool use) in the context of foundation models?

A A structured mechanism where the model outputs a specific function name and arguments that the application then executes, returning the result to the model B Calling one foundation model from within another C The model executing Python code directly D Using AWS Lambda to invoke the model

40 / 60

Question 41 of 60

A RAG application retrieves 10 chunks but only 2 are actually relevant to the query. This degrades response quality by flooding the context window with irrelevant information. Which retrieval metric measures this problem?

A Answer Relevancy B Context Precision C Faithfulness D Context Recall

41 / 60

Question 42 of 60

Which Amazon Bedrock capability allows developers to define and version reusable prompt templates with variables that can be filled at runtime?

A Bedrock Agents B Bedrock Prompt Management C Bedrock Knowledge Bases D Bedrock Fine-tuning

42 / 60

Question 43 of 60

A developer is building a document Q&A system. After retrieval, they want to re-rank the retrieved chunks to surface the most relevant ones before passing them to the LLM. What component performs this function?

A Tokenizer B Vector database C Reranker model D Embedding model

43 / 60

Question 44 of 60

What is the purpose of 'guardrail trace' information returned by Amazon Bedrock Guardrails?

A To provide model reasoning steps B To log inference latency metrics C To display the retrieved RAG chunks D To show which guardrail policies were evaluated and whether they triggered, for debugging

44 / 60

Question 45 of 60

A company has strict data residency requirements and cannot allow their prompts or data to leave a specific AWS region. How does Amazon Bedrock address this?

A Bedrock stores all data in us-east-1 by default B Data residency is not supported by managed AI services C Bedrock processes data within the AWS region where the API is called, and cross-region inference must be explicitly configured D Bedrock automatically routes requests globally for lower latency

45 / 60

Question 46 of 60

Which Bedrock Agents concept defines the specific API operations an agent is allowed to call, specified via an OpenAPI schema?

A Memory B Knowledge Base C Action Group D Orchestration prompt

46 / 60

Question 47 of 60

A developer is fine-tuning a foundation model on Amazon Bedrock and wants to prevent the model from losing its general capabilities while adapting to a new domain. Which fine-tuning technique is designed to address this?

A Continued pre-training B Full fine-tuning with a large learning rate C Continued Fine-Tuning with low learning rate and small dataset D LoRA (Low-Rank Adaptation) fine-tuning

47 / 60

Question 48 of 60

What is the correct sequence of steps in a RAG pipeline when a user submits a query?

A Store â†’ Chunk â†’ Embed â†’ Retrieve â†’ Augment â†’ Generate B Query â†’ Embed query â†’ Retrieve similar chunks â†’ Augment prompt â†’ Generate response C Query â†’ Fine-tune model â†’ Generate response D Query â†’ Generate response â†’ Retrieve â†’ Validate

48 / 60

Question 49 of 60

A developer wants their Bedrock agent to maintain context about user preferences across multiple separate conversations over days. Which feature enables this persistent memory?

A Bedrock Agents memory (long-term memory) B Guardrails C Session attributes (in-session only) D Bedrock Knowledge Bases

49 / 60

Question 50 of 60

Which AWS service should be used to monitor Amazon Bedrock application performance metrics such as invocation latency, token counts, and error rates?

A AWS Config B AWS Trusted Advisor C Amazon CloudWatch D AWS CloudTrail

50 / 60

Question 51 of 60

A developer's RAG application retrieves the correct chunks but the LLM still generates wrong answers. The issue is that the LLM is ignoring the retrieved context and using its pre-trained knowledge instead. What is the best mitigation?

A Reduce the temperature parameter B Explicitly instruct the model in the system prompt to answer ONLY based on the provided context and to say it doesn't know if the answer is not in the context C Use a larger embedding model D Retrieve more chunks

51 / 60

Question 52 of 60

Which foundation model family available on Amazon Bedrock is developed by Amazon and includes models for text generation and embeddings?

A Claude B Llama C Mistral D Amazon Titan

52 / 60

Question 53 of 60

A developer needs to select a foundation model that can accept an image and a text question as input and generate a text answer about the image. What model capability is required?

A Extended context window B Streaming inference C Multimodal input (vision + text) D Text-to-image generation

53 / 60

Question 54 of 60

What is 'top-k' sampling and how does it differ from top-p?

A Top-k limits the maximum tokens generated; top-p limits the minimum tokens B Top-k and top-p are identical parameters with different names C Top-k sets the temperature; top-p sets the diversity D Top-k restricts sampling to the k highest-probability tokens; top-p dynamically includes tokens until the cumulative probability reaches p

54 / 60

Question 55 of 60

A company's Bedrock application is returning inconsistent results for the same prompt. The developer wants to make the outputs fully deterministic for reproducibility. What should they set?

A Max tokens to a fixed value B Top-p to 0.5 C Temperature to 0 (and top-k to 1 if supported) D Temperature to 0 and top-p to 1

55 / 60

Question 56 of 60

In Amazon Bedrock's agent architecture, what is the 'orchestration prompt' (or 'instructions' field)?

A The OpenAPI schema for action groups B The prompt used to query the knowledge base C The user's first message to the agent D The system-level instructions that define the agent's persona, capabilities, and reasoning strategy

56 / 60

Question 57 of 60

A developer is implementing security for a multi-tenant Bedrock application where different tenants must not see each other's data in the shared knowledge base. What is the recommended architecture?

A Create separate knowledge bases per tenant with separate IAM roles B Store all tenant data in a single S3 bucket with encryption C Use Guardrails to separate tenant data D Use a single knowledge base with tenant filtering via metadata

57 / 60

Question 58 of 60

What is the risk of setting max_tokens too low in a production Bedrock application?

A Responses will be truncated mid-sentence, degrading user experience B The model will hallucinate more C API costs will increase D The model will use more compute

58 / 60

Question 59 of 60

A developer is building a document ingestion pipeline for Bedrock Knowledge Bases. The source documents contain many tables and structured data. Which chunking approach is most appropriate?

A Hierarchical chunking or custom chunking that respects table boundaries B Fixed-size character chunking C Semantic chunking D Random chunking

59 / 60

Question 60 of 60

Which responsible AI practice should a developer implement to detect when a Bedrock application's output quality degrades over time due to data drift or model updates?

A Deploy guardrails and consider the problem solved B Ask users to manually flag incorrect responses C Only test the application before initial deployment D Implement continuous evaluation pipelines that regularly sample and score production outputs using metrics like faithfulness and answer relevancy

60 / 60

AWS Generative AI Developer AIP-C01 Practice Quiz