Task 1.5: Design Retrieval Mechanisms for FM Augmentation. AIP-C01 Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A financial services company is building a RAG system for their knowledge base. Users often ask vague questions like 'What are the risks of investing in emerging markets?' The knowledge base contains many documents, each covering multiple topics. They want to improve retrieval effectiveness. What is the best approach?

   • AUse Step Functions to transform the query into a structured format before searching.
   • BSimply use the original query as-is and rely on the vector search to find relevant documents.
   • CUse Amazon Bedrock to expand the query into multiple specific sub-queries and retrieve documents for each.
   • DUse a Lambda function to decompose the query into smaller parts and then search for each part.
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   A developer wants to enable a foundation model to call a retrieval function to fetch relevant documents. Which interface allows the FM to invoke external tools like a vector search API?

   • AStandardized API patterns
   • BModel Context Protocol (MCP) client
   • CVector search API
   • DFunction calling interface
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   A developer is building a search system that combines traditional keyword-based search with semantic vector search to improve retrieval relevance. Which architecture best describes this approach?

   • AKeyword-only search that ignores semantic relationships.
   • BReranking with a cross-encoder model applied to the top results from a single search method.
   • CHybrid search that combines keyword and vector search results using a fusion algorithm.
   • DPure vector search that uses only embeddings to find similar documents.
   Reveal Solution Discussion Select an answer first
5. 24expert · hard

   A company is building a RAG system over a large corpus of technical manuals. Each manual is a PDF with a table of contents, chapters, and sections. They need to retrieve precise troubleshooting steps (e.g., 'How to fix error code 1234?') and also answer conceptual questions (e.g., 'What is the architecture of the system?'). They are using Amazon Bedrock Knowledge Bases with OpenSearch Serverless. They have a constraint: the embedding model has a maximum input token limit of 512 tokens. What is the most effective chunking strategy?

   • AUse fixed-size chunking with 1024 tokens and rely on the model to truncate.
   • BUse fixed-size chunking with 512 tokens and no overlap.
   • CUse a single chunk per chapter, even if it exceeds the token limit.
   • DUse hierarchical chunking with parent chunks at the chapter level and child chunks at the section level, ensuring child chunks are under 512 tokens.
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   A news organization is building a search system over their articles. They want to support both exact phrase queries (e.g., 'climate change') and semantic queries (e.g., 'global warming effects'). They are using Amazon OpenSearch Service. What is the best way to configure the search?

   • AUse a single index with both keyword and vector fields, but only query the vector field.
   • BConfigure a hybrid search with both a keyword index and a vector index, and use a search pipeline to combine results.
   • CUse only a keyword search with a custom analyzer for exact phrase matching.
   • DUse only a vector search with embeddings from Amazon Titan.
   Reveal Solution Discussion Select an answer first