🔗 Don't bother parsing: just use images for RAG

A visual approach to document retrieval, using images of pages instead of traditional OCR and parsing methods, is a possible method to improve accuracy and preserve context for document parsing. This company divides document images into patches and uses vision transformers and language models to understand both textual and visual elements. By treating documents as visual objects, they are able to have higher accuracy in document understanding and search compared to traditional methods, especially for complex documents with charts, tables, and diagrams.