If your RAG pipeline answers questions about PDFs badly, look at the chunks before you tune the embeddings or the prompt. A lot of retrieval problems start at extraction: two-column pages read straigh
Markdown supported · ⌘↵ to post
No responses yet.