Home / GenAI Engineer practice test / Application Development / Question

Databricks Generative AI Engineer Associate · free practice question

A data scientist is comparing two text segmentation approaches for a Retrieval-Augmented Generation (RAG) system handling medical literature. The first…

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

Updated November 11, 2026

Application Development

A data scientist is comparing two text segmentation approaches for a Retrieval-Augmented Generation (RAG) system handling medical literature. The first approach employs fixed-size recursive character splits (512 tokens with 50-token overlap). The second approach utilizes semantic chunking, aggregating sentences based on their embedding similarity. While the semantic approach yields superior Mean Reciprocal Rank (MRR) during retrieval, it demands considerably more time for initial indexing. When would the semantic chunking approach (Strategy B) be the more suitable option?

  1. AWhen the document collection is updated frequently and fast index generation is a critical requirement
  2. BWhen documents have long, dense paragraphs that fixed-size splits would cut mid-idea
  3. CWhen evaluation queries mostly seek highly precise answers that are contained in a single sentence
  4. DWhen the embedding model has a small context window, under 256 tokens, and cannot encode complete paragraphs
Answer & explanation

Correct answer: B

WHY B is correct: Semantic chunking groups sentences by embedding similarity, preserving topical coherence independent of character count. When medical research papers contain dense multi-sentence paragraphs that discuss a single concept, semantic chunking keeps related sentences together — yielding higher MRR because each retrieved chunk fully encapsulates the relevant context.

WHY NOT A: Near-real-time corpus updates favor Strategy A (recursive splitting) because it is computationally cheap. Strategy B's higher indexing time makes it unsuitable for high-frequency updates.

WHY NOT C: Queries requiring precise single-sentence answers benefit from more granular, fixed-size chunks (Strategy A). Semantic chunking groups multiple sentences together, diluting precision for single-sentence lookups.

WHY NOT D: A short-context-window embedding model struggles with larger semantic chunks. Strategy B would produce chunks that exceed a 256-token model's window, causing truncation and degraded embeddings.

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 →