OCR-free document-understanding transformer (PyTorch/Hugging Face) that classifies document types, extracts structured information (e.g. from receipts), answers questions about document content, and reads text sequences directly from document images without a separate OCR engine.