AWS Certified Generative AI Developer Professional AIP-C01 Practice Questions with Explan…

155 questions · Cloud Computing · by Study Foxy

Study these questions free

Questions

  1. A support assistant must answer from 50,000 internal articles that change daily, cite sources, and avoid retraining. Which design fits?

    • RAG over a managed knowledge base
    • Monthly full fine-tuning on all articles
    • Pre-training a new model from scratch
    • Prompting a model with no retrieval at all

  2. A team must quickly prove whether a GenAI idea delivers value before investing in production. What should come first?

    • A small proof of concept with clear success metrics
    • A full multi-Region production rollout
    • Fine-tuning three models in parallel
    • Buying a year of Provisioned Throughput

  3. Which Well-Architected resource offers GenAI-specific guidance on design trade-offs?

    • The Generative AI Lens
    • The SaaS Lens
    • The IoT Lens
    • The Games Industry Lens

  4. A design must balance answer quality, latency under 2 seconds, and cost per request. What should drive model choice?

    • Tests on the team's own tasks and targets
    • The model with the most parameters available
    • Whichever model was released most recently
    • The model with the longest name in the catalog

  5. Which requirement most strongly suggests RAG over fine-tuning?

    • Answers must reflect knowledge that changes every week
    • Output must follow a fixed, specialized brand voice
    • The model must learn a new structured output format
    • Responses must use a domain's specialized writing style

  6. Which requirement most strongly suggests fine-tuning?

    • A specialized style prompting can't reach
    • Answers must cite today's policy documents
    • Data changes every hour across sources
    • The team has no labeled examples at all

  7. Which statement about designing for model portability is correct?

    • Abstract calls behind a unified interface like Converse
    • Hardcode one provider's request format everywhere
    • Avoid any abstraction to save a few milliseconds
    • Store prompts inside compiled application binaries

  8. An app must call several Bedrock models with one consistent request and response format. Which API fits?

    • The Bedrock Converse API
    • Model-specific InvokeModel bodies
    • The Bedrock control plane API
    • The SageMaker training API

  9. A model must be swapped for a newer version without code changes. Which design fits?

    • Keep the model ID in AppConfig
    • Hardcode the model ID in every function
    • Ask users to choose the model each time
    • Redeploy the whole stack for every change

  10. Bedrock requests are throttled during regional peaks. Which feature spreads load across Regions?

    • A cross-Region inference profile for the geography
    • Provisioned Throughput in one Region only
    • A larger max_tokens value on each request
    • Batch inference jobs scheduled nightly

  11. A workload needs guaranteed, consistent throughput for a model. Which option fits?

    • Provisioned Throughput
    • On-demand inference with no reservation
    • Batch inference with a daily schedule
    • Cross-Region inference with a global profile

  12. A team fine-tuned an open-weight model elsewhere and wants to serve it through Bedrock APIs. Which feature fits?

    • Bedrock Custom Model Import of the weights
    • Bedrock continued pre-training on the base model
    • Bedrock Knowledge Bases with a custom parser
    • Bedrock Marketplace model subscriptions

  13. A smaller, faster model should be trained to imitate a larger model for one task. Which Bedrock feature fits?

    • Model distillation
    • Continued pre-training
    • Prompt caching
    • Cross-Region inference

  14. Which inference parameter limits sampling to the most likely tokens whose probabilities add up to a threshold?

    • Top-p
    • Max tokens
    • Stop sequences
    • Temperature only

  15. A model must stop generating as soon as it outputs the string ###. Which parameter fits?

    • A stop sequence
    • Top-k
    • Temperature
    • A system prompt

  16. Which model type converts text into vectors for semantic search?

    • An embedding model
    • An image generation model
    • A speech synthesis model
    • A text-to-SQL model

  17. A domain has lots of unlabeled text, and the model needs better domain vocabulary. Which customization fits?

    • Continued pre-training
    • Instruction fine-tuning only
    • Prompt caching
    • Raising temperature

  18. Lowering temperature makes a model's outputs more deterministic.

    True or false

  19. Scanned PDFs with tables must be parsed into clean text before indexing. Which option fits?

    • Advanced FM-based parsing
    • Default text parsing only
    • Uploading the PDFs as images to S3
    • Converting the PDFs to ZIP files

  20. Documents must be checked for missing fields and duplicates before they feed a model. Which service fits?

    • AWS Glue Data Quality
    • Amazon Bedrock Guardrails
    • Amazon Kendra connectors
    • AWS Glue Schema Registry checks

  21. Customer emails must be stripped of PII before being sent to a model. Which approach fits?

    • Detect and redact PII in a preprocessing step
    • Send raw emails and filter outputs only
    • Encrypt emails and send them anyway
    • Ask customers not to include PII

  22. Images, audio, and text from one claim must be processed together. Which capability fits?

    • A multimodal pipeline with suitable models
    • A text-only tokenizer for everything
    • A relational database trigger
    • A CloudFront cache behavior

  23. Large batches of documents must be enriched by a model asynchronously at lower cost. Which option fits?

    • Bedrock batch inference
    • Real-time streaming calls
    • Provisioned Throughput for an hour
    • A larger context window

  24. Model inputs must follow a fixed JSON schema before submission. Where should validation happen?

    • In a preprocessing step before the call
    • Only after the model responds
    • Nowhere, since models fix bad input
    • Inside the model's system prompt only

  25. A team needs a serverless vector store with hybrid search for a Bedrock knowledge base. Which fits?

    • Amazon OpenSearch Serverless
    • Amazon DynamoDB with TTL
    • Amazon ElastiCache without vectors
    • Amazon Redshift Spectrum

  26. A company wants the lowest-cost vector storage for a large, infrequently queried embedding corpus. Which option fits?

    • Amazon S3 Vectors
    • OpenSearch with provisioned nodes
    • An in-memory cache cluster
    • Provisioned IOPS EBS volumes

  27. An app already uses Aurora PostgreSQL and wants vectors next to relational data. What should it use?

    • The pgvector extension in the existing Aurora cluster
    • Aurora zero-ETL integration to Redshift
    • An Aurora Global Database secondary Region
    • Babelfish for Aurora PostgreSQL

  28. Retrieval must return only documents a user's department can see. Which feature fits?

    • Metadata filters on the user's department
    • A larger embedding dimension for all chunks
    • A Guardrails denied topic per department
    • A separate model for each department

  29. Which index type offers fast, high-recall approximate nearest-neighbor search at higher memory cost?

    • HNSW
    • B-tree
    • Bitmap
    • Hash

  30. Documents change daily in S3. How should the vector index stay current?

    • Incremental data source syncs
    • Rebuild the model every day
    • Delete the index every hour
    • Store vectors only in memory

  31. Titan Text Embeddings V2 lets you choose smaller vector sizes. What is the trade-off?

    • Less storage, some accuracy risk
    • Higher accuracy and higher cost
    • No effect on storage or cost
    • Faster training of the FM itself

  32. Long documents lose context when split into small chunks. Which chunking strategy keeps both detail and context?

    • Hierarchical chunking
    • No chunking at all
    • Fixed 50-token chunks
    • Random chunk sizes

  33. Chunks often break mid-idea, hurting relevance. Which strategy splits by meaning?

    • Semantic chunking
    • Fixed-size chunking
    • Character chunking
    • No overlap chunking

  34. Exact product codes are missed by vector search alone. What helps?

    • Hybrid search
    • Higher temperature
    • Smaller chunks only
    • Removing metadata

  35. A complex question needs facts from several documents. What improves retrieval?

    • Query decomposition
    • Retrieving only one result
    • Disabling reranking
    • Using a shorter prompt

  36. An app needs retrieved passages only, to build its own prompt. Which Knowledge Bases API fits?

    • Retrieve
    • RetrieveAndGenerate
    • InvokeAgent
    • CreateDataSource

  37. Relevant passages are retrieved but ranked poorly. What should be added?

    • A reranker model
    • A larger model only
    • More chunk overlap
    • A higher top-p

  38. A team needs enterprise search over SharePoint and Confluence with built-in connectors as a RAG retriever. Which fits?

    • Amazon Kendra
    • Amazon Athena
    • AWS Glue DataBrew
    • Amazon MQ

  39. Prompts must be versioned, tested, and reused across apps with variables. Which feature fits?

    • Bedrock Prompt Management
    • Bedrock Guardrails versions
    • Bedrock batch inference jobs
    • Parameter Store with no testing

  40. A model must always return valid JSON for downstream parsing. What helps most?

    • Define a schema with tool use
    • Raise the temperature
    • Ask it to be creative
    • Remove the system prompt

  41. Instructions about tone and rules should apply to every turn of a conversation. Where should they go?

    • The system prompt
    • The first user message only
    • The stop sequences
    • The model's ID

  42. A model must reason through multi-step math before answering. Which technique helps?

    • Chain-of-thought prompting
    • Zero-shot prompting with lower top-p
    • Shorter prompts with no instructions
    • Negative prompting about math errors

  43. Several prompts must run in a fixed sequence, passing outputs along, with no custom code. Which feature fits?

    • Bedrock Flows
    • Bedrock Guardrails
    • S3 Lifecycle rules
    • AWS Budgets

  44. Who should be able to change production prompts?

    • Approved roles, with review
    • Anyone with console access
    • Every end user of the app
    • No one, ever, after launch

  45. Few-shot examples in a prompt can improve output consistency.

    True or false

  46. Which store suits a knowledge graph plus vector search for connected entities in RAG?

    • Amazon Neptune Analytics
    • Amazon S3 Glacier storage
    • Amazon Kinesis Data Streams
    • AWS Glue DataBrew recipes

  47. Larger chunks always improve retrieval accuracy.

    True or false

  48. A model needs domain style changes from 5,000 labeled prompt-response pairs. Which customization fits?

    • Fine-tuning the model
    • Continued pre-training only
    • Prompt caching only
    • Retrieval with no examples

  49. A Bedrock agent must call an internal order API. How is the API exposed to the agent?

    • An action group with an OpenAPI schema
    • A Guardrails denied topic list
    • A Knowledge Base data source
    • A Prompt Management variable

  50. An agent should hand control back to the app to execute an action itself, such as a local transaction. Which feature fits?

    • Return of control
    • Prompt caching
    • Cross-Region inference
    • Batch inference

Showing 50 of 155. Open the quiz to study them all.