An approach from #NLP to provide useful chunks of e. g. texts, useful in combination with generating embeddings.

A tool for chunking is the langchain text splitter: https://reference.langchain.com/python/langchain_text_splitters/

Chunking with overlap

For Embeddings & vectorization based Retrieval augumented generation (RAG) systems, chunking with an overlap can help to improve quality. Yet, the downside is that the system might find the same original sentence but as two, split up documents in the top-5 nearest documents. Therefore, if the lookup was a bad shot anyways, the second nearest document is useless to.