Návrh a prototyp AI Agenta pro generování datových modelů

This thesis presents the design, implementation, and experimental evaluation of an AI agent prototype for assisted data modeling. The work is grounded in the premise that large language models cannot currently be considered sufficiently reliable for fully autonomous database schema design and therefore proposes an alternative approach — an iterative workflow with active user involvement through a Human-in-the-Loop mechanism. The theoretical part identifies key limitations of generative AI in conceptual and semantic modeling and establishes the principles underlying the proposed solution. The practical part describes the agent's architecture, its technological foundations, and the execution of two testing scenarios — processing structured CSV data and generating a data model from a natural language business specification. Experimental evaluation demonstrates that the agent can produce structurally sound conceptual drafts, with quality improving significantly after HITL corrections. The results confirm that an iterative approach with guided human validation is more suitable than one-shot autonomous generation, particularly for complex domains. The thesis contributes to a shift from the paradigm of full automation toward controlled, human-assisted data modeling.