Customize Splitter for specific scenarios

[Jolmoz]

Thank you for this awesome project!

In my practical document processing experience, I sometimes encounter cases where a PDF contains multiple invoices, some single-page and others multi-page.

In my case, the TextSplitter worked well when dividing a group of documents with different classifications, but when the documents have the same classification but are different documents, the separator groups them as a single document.

For example, here are two invoices (same classification, but different documents in the same PDF), and the separator groups them as the same document, pages 1 and 2.

I was reading a bit the code and I found the instruction to the LLM

I was wondering if it might be possible to add a parameter to add or modify that instruction to tell the LLM about cases like the ones I just mentioned.