Played with RAG a bit, quite nice. Now looking for a similar solution that generates pair-completion pairs in flat json from MD or TXT files, for prompt augmentation (fine tuning) purposes. Retrieval does not meet 100% my need.
Normally the “extract from file” node in n8n should give you plenty of metadata, certainly it does for pdfs. Though strangely I didn’t find the file name in the metadata of my pdf either. The secondary option would be to pass it through a code node to add the name of the binary file to the output json.
That’s awesome, how the heck did you get the OAuth to work? It didn’t like my redirect URI and I could only run it in test mode, which isn’t a huge deal but just curious how you did it?