Fixed-size chunking splits documents into segments with a predetermined length. For semantic search, similar-length chunks help balance two risks: short segments may lack enough meaning to match queries, while long segments may mix topics and produce noisy embeddings. section 7.2.2
Chunking lengthy documents
A collection of detailed reports and unstructured notes is split into 300-token chunks. The chunks are then embedded with MiniLM and stored in Weaviate for semantic search. section 7.2.2
