Usually retrieval system then fetch each chunk individually. The idea behind this is you fetch the parent document they come from (for more context)

→ View original post on X — @hwchase17