Apply a Chunking Strategy for a Given Document Structure and Model Constraints GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 4)

Questions 16–19

1. 16expert · hard

   A team is building a RAG system for a collection of scientific papers. The papers have a standard structure: abstract, introduction, methods, results, and discussion. The team has implemented a structure-based chunking strategy that splits on these sections. During evaluation, they find that the system performs well on questions about the 'methods' section but poorly on questions that require synthesizing information from both the 'results' and 'discussion' sections. What is the most likely cause and the best solution?

   • AThe chunking strategy is not preserving the relationship between the 'results' and 'discussion' sections. The solution is to add an overlap or merge these sections into a single chunk.
   • BThe chunks are too large, so the retrieval system is returning too much irrelevant information. The solution is to reduce the chunk size.
   • CThe chunks are too small, so the retrieval system is not returning enough context. The solution is to increase the chunk size.
   • DThe embedding model is not capable of understanding the relationship between the sections. The solution is to use a different embedding model.
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A team is chunking a series of long-form interview transcripts for a knowledge base. The transcripts have a conversational structure with questions and answers. The target model has a context window of 2048 tokens. They want to ensure that a question and its corresponding answer are always in the same chunk. What is the most important parameter to configure?

   • AThe overlap, setting it to a value large enough to cover the longest possible answer
   • BThe chunk boundaries, ensuring they are set to break at the start of a new question
   • CThe chunk size, setting it to exactly 2048 tokens
   • DThe embedding model, choosing one that is optimized for conversational text
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A team is building a RAG system for a collection of product manuals. The manuals have a consistent structure: a table of contents, chapters, sections, and step-by-step instructions. They want to answer questions like 'How do I reset the device?' and 'What are the safety warnings?'. Which document structure analysis is most important for informing their chunking strategy?

   • AIdentifying the images and diagrams to ensure they are included in the chunks
   • BIdentifying the font size and style of the text to determine the importance of each section
   • CIdentifying the headings and subheadings to define the hierarchy of topics
   • DIdentifying the page numbers of each section to ensure chunks are aligned with pages
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   A document contains long, unstructured paragraphs with no clear headings. Which chunking strategy is most appropriate to preserve semantic meaning?

   • ASemantic chunking based on embedding similarity
   • BStructure-based chunking using HTML tags
   • CFixed-size chunking with a token count of 100
   • DRecursive character text splitting
   Reveal Solution Discussion Select an answer first