Teaching PyTorch To Read Your Worst PDFs With Docling - Mingxuan Zhao, Peter Staar & Carol Chen

This talk focuses on the challenges of extracting clean, structured data from real-world documents, particularly PDFs, which often result in lost information due to formatting issues. The speaker introduces Docling, an open-source document parsing library that employs PyTorch-based deep learning models to analyze documents as humans do, maintaining hierarchy and extracting structured data from various formats. The session explores Docling's architecture and demonstrates how to build document processing pipelines for retrieval-augmented generation (RAG) and other generative AI applications. Attendees will gain insights into integrating Docling with their own PyTorch workflows, enhancing their data processing capabilities.