179 questions
No questions match those filters.
← All questions
RAGMedium· Chunking· Evaluation· Data Analysis
Your chunks are 512 fixed tokens and recall@10 is 0.71....
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansA strong answer refuses to speculate without understanding the corpus. Semantic chunking relies on a cosine detector’s ability to identify boundaries. If the corpus has high-delta boundary gaps (e.g., 0.90) compared to within-topic gaps (e.g., 0.25), semantic chunking may work. However, if the document structure results in low-delta boundaries (e.g., 0.35), the detector will fail, potentially causing a significant drop in recall compared to fixed-size chunking. Simply claiming semantic chunks ‘preserve meaning’ is insufficient.