Have you tried NotebookLM which basically does this as an app on the bg (chunking and summarising many docs) and you can -chat- with the full corpus using RAG