AIF-C01 Practice Question 43: Document Chunking
AWS Certified AI Practitioner (AIF-C01) · Test 5 · question 3 of 10
A company is loading hundreds of long policy manuals into an Amazon Bedrock knowledge base. During ingestion, each manual is split into smaller pieces, and each piece is converted into an embedding and stored in a vector index with a link back to its source document. At query time, only the most relevant pieces are added to the prompt. What is this splitting step called?
Show answer and explanation
Answer:
Chunking splits documents into manageable chunks for efficient retrieval. In Amazon Bedrock Knowledge Bases, the chunks are converted to embeddings and written to a vector index while keeping a mapping to the original document. That lets a query retrieve only the relevant passages instead of whole manuals (option D). Bedrock supports several strategies: default chunking (about 300 tokens, honoring sentence boundaries), fixed-size chunking with overlap, hierarchical chunking, semantic chunking, or no chunking.
- A is wrong because tokenization breaks text into *tokens*, the small units a model processes such as words or parts of words. It isn't how documents are divided into retrievable passages. - B is wrong because fine-tuning trains a model on labeled examples to change its behavior. The documents here aren't used to change the model at all. - C is wrong because distillation transfers knowledge from a larger *model* to a smaller one. It has nothing to do with splitting documents.