Home / GenAI Engineer practice test / Data Preparation / Question

Databricks Generative AI Engineer Associate · free practice question

An AI solutions architect is developing a Retrieval Augmented Generation (RAG) system to process a vast archive of scientific articles. Each article comprises…

Exam-style question on Data Preparation from the free GenAI Engineer practice test, with the answer and a full explanation below.

Updated November 11, 2026

Data Preparation

An AI solutions architect is developing a Retrieval Augmented Generation (RAG) system to process a vast archive of scientific articles. Each article comprises a title, abstract, several numbered sections with headings, and a bibliography. Users will pose questions ranging from highly specific, factual queries about an individual paper's abstract (e.g., 'What was the key finding reported in the abstract of Miller et al. (2022) regarding quantum entanglement?') to broader, conceptual inquiries spanning multiple publications (e.g., 'Discuss the prevailing viewpoints on carbon capture technologies across recent literature.'). The architect's objective is to build a single RAG system capable of efficiently handling both granular, source-specific information retrieval and wide-ranging, thematic discovery. What design approach for the retrieval system offers the MOST effective support for these diverse querying needs?

  1. AChunk abstracts separately from body paragraphs, tag chunks with section and paper metadata, and filter on it for paper-specific queries
  2. BSplit each article into uniform 512-token segments with a small overlap and search the whole index for both query types
  3. CUse an LLM to break each sentence into atomic factual propositions and index each proposition as its own retrieval unit
  4. DIndex the abstract of each paper as a single chunk and leave out the body, on the basis that abstracts summarize the key findings
  5. ERandomly sample 200-token snippets from each paper and index those, to get unbiased coverage across the paper's sections
Answer & explanation

Correct answer: A

WHY A: Metadata-enriched chunking with semantic boundaries combines two techniques that suit academic papers. Section-aware metadata: tagging each chunk with section: 'abstract', paper_id, year, authors enables metadata filtering at retrieval time, so for the Miller et al. (2022) abstract question the retriever can filter WHERE section = 'abstract' AND authors CONTAINS 'Miller' AND year = 2022 before similarity search, reducing the search space and improving precision. Semantic boundaries: splitting body sections at paragraph boundaries rather than arbitrary token positions preserves the coherent argument structure of academic paragraphs, improving embedding quality. Dual query support: for broad conceptual queries (such as the carbon capture question), a full index search with no metadata filter retrieves the most semantically relevant paragraphs from across all papers.

WHY NOT B: Uniform 512-token chunks ignore the structural signal that the 'abstract' section provides. A query asking about an abstract will retrieve any chunk, not necessarily one from the abstract. Structured documents with distinct section types benefit from structure-aware chunking.

WHY NOT C: Proposition-based chunking decomposes content into atomic factual claims and works well for dense factual retrieval. However, it requires running an LLM over each sentence of each paper (computationally expensive for a large academic corpus), and propositions lose the relational context between claims within a paragraph, which matters for synthesis queries about debates and frameworks.

WHY NOT D: Indexing abstracts alone discards 90%+ of each paper's informational content. The body sections contain methodology, experimental results, and discussion of debates that are essential for synthesis questions. An abstract-only index cannot answer the carbon capture question because the competing viewpoints are in the discussion sections, not the abstract summaries.

WHY NOT E: Random subsampling introduces arbitrary coverage gaps. Critical sections may be underrepresented or absent, and there is no principled way to ensure that specific sections (abstracts, conclusions) or specific arguments are consistently represented. It trades a solvable structural chunking problem for non-deterministic coverage quality.

Take the full GenAI Engineer practice test free →

Would you pass today?

This question is one of 45 in the free, timed Databricks Generative AI Engineer Associate practice test. Take it and get a section-by-section score.

Start the free test →