DeepSeek OCR - Extract text with 97% accuracy using a vision-language model | PeerPush