Only Vespa allows you to index multiple vectors per schema field, avoiding duplicating all the meta data of the document into the "chunk", and avoids maintaining the document to chunk fan-out. See https://blog.vespa.ai/semantic-search-with-multi-vector-inde...