The DocLap interface for document layout analysis and interaction

We present DocLap, a demo system for interactive exploration and semantic analysis of scholarly documents. The system integrates document layout analysis, semantic enrichment, and Large Language Model (LLM) interoperability in a unified framework, allowing multilayered access to document content. DocLap processes both PDF and LaTeX sources, extracting structural and visual elements such as sections, figures, tables, and captions. The interactive GUI supports segmentation and inspection of layout components through dynamic pop-ups, allowing users to navigate and validate extracted elements. Visual information is linked to semantic data, enabling ontology-based constraint violation detection to identify layout inconsistencies. LLMs are employed for answering questions across the whole document and for question classification. The framework also includes a ground truth builder for bounding box annotation.