Conceptual modeling, particularly with Entity-Relationship Diagrams (ERDs), is traditionally a manual and complex task in data engineering. This paper presents a method and tool for automating ERD generation from semi-structured data using Large Language Models (LLMs). The approach follows an iterative cycle, data ingestion, prompt design and execution, and results evaluation, with human-in-the-loop refinement to improve accuracy through expert feedback. Validation was performed on two datasets, assessing model quality via coverage, referential integrity, and Third Normal Form (3NF) compliance. Results showed high scores (0.90 and 0.93), confirming the tool’s effectiveness and the need for expert input to resolve complex modeling issues. This work offers an LLM-assisted solution to assist in complex data engineering tasks, while acknowledging the continued importance of expert oversight.