With the latest update, Pinecone can ingest PDFs straigth away. What parsing/extracting does Pinecone use there? I have difficult-to-handle PDFs and want to learn more about how well Pinecone parses them before fully switching to Pinecone. Does anybody know?

Hi! Welcome to the Discord. Have you tried assistant yet with these docs? I'm not sure if we have public-facing docs on our extraction strategy, but you can throw up to 100 pdfs into an assistant for free and try it yourself to see what's going on You can also use the context API to see the chunks being returned/referred to

Thanks Arjun! I tried this and it worked well on simpler PDFs and does poorly on the ones I expected it to have a more challenging time. Won't be able to make the switch quite yet, but perhaps it would make sense to open up the PDF chunking and parsing processes and garner community feedback. Unfortunately PDFs are still a pain to work with.

Are your PDFs multimodal, by any chance?

@plain-winter-59357 can you share any of the PDFs and queries it's struggling with so we can analyse them in more depth?