Late Chunking: Balancing Precision and Cost in Long Context Retrieval

this is like performing RAG after the query pipeline is done? I'm not sure I understand the mechanisms at work here...