An enterprise application uses an Amazon Bedrock foundation model (FM) to process and analyze 50 to 200 pages of technical documents. Users are experiencing inconsistent responses and receiving…

Question

Options

• AConfigure fixed-size chunking at 4,000 tokens for each chunk with 20% overlap. Use application-
• BUse hierarchical chunking with parent chunks of 8,000 tokens and child chunks of 2,000 tokens.
• CUse semantic chunking with a breakpoint percentile threshold of 95% and a buffer size of 3
• DCreate a pre-processing AWS Lambda function that analyzes document token count by using the

• A
  12% (3)
• B
  24% (6)
• C
  60% (15)
• D
  4% (1)

Explanation

Semantic chunking splits documents at natural meaning boundaries rather than at arbitrary token counts. A breakpoint percentile threshold of 95% means chunks are only split when there is a very large semantic shift, keeping closely related content together within each chunk. This directly addresses inconsistent responses (caused by splitting semantically related content across chunks) and truncation (caused by poor chunk boundaries that cut off reasoning mid-thought). Option A (fixed-size chunking) is mechanical and ignores content meaning, leading to the same inconsistency problems. Option B (hierarchical chunking) is better than fixed-size but still relies on token-count limits rather than semantic coherence. Option D (Lambda pre-processing to analyze token count) only detects when a document is too long; it does not solve how to intelligently split it for coherent retrieval.

No community discussion yet for this question.