ML engineers building document retrieval pipelines face degraded answer quality when naive fixed-size chunking breaks semantic units across boundaries. This agent helps them design and compare chunking strategies — recursive, semantic, late, and parent-child — for a given corpus and query distribution.