Generative Ontology Induction: Domain-Agnostic Schema Discovery from Document Corpora Using Large Language Models

This work presents an ontology-driven approach to structuring AI systems, focusing on bridging the gap between research and engineering practices. While modern AI systems achieve strong empirical performance, they often lack explicit structural representations that enable interpretability, consistency, and cross-team collaboration. We introduce a framework for constructing and managing ontologies as a foundational layer for AI development. The proposed approach enables teams to formalize domain knowledge, standardize system components, and improve alignment between research concepts and production systems. The framework is implemented as a practical tool that allows users to create, visualize, and refine ontologies from structured and semi-structured inputs. The system supports both manual ontology design and automated generation from example datasets, providing a flexible workflow for different stages of development. We demonstrate how this approach can serve as a structural backbone for AI systems, improving maintainability, scalability, and communication across multidisciplinary teams. This work is presented in the context of ongoing research on ontology-based AI systems and is aligned with discussions at ESWC on integrating semantic technologies with modern machine learning workflows.