Chunk-level Retrieval

A granularity strategy in document retrieval that splits long documents into passages of roughly a few hundred characters, each vectorized and indexed separately. Compared with whole-document embedding, it improves recall precision and lets results point directly to specific content, but it multiplies the number of index entries and increases capacity planning pressure. Splitting strategies include fixed length, heading-based and semantic boundary-based approaches.