Leveraging Llm In The Implementation of The Cfihos Standards Conceptual Model

This work presents the experience of using Large Language Models (LLMs), specifically GPT-4o, as a tool to support part of the development and testing of tools related to CFIHOS - a standard for industrial data handover process. The LLM is used to generate SQL DDL schemas from entity-relationship descriptions in both textual and spreadsheet formats. Additionally, we implemented the physical schema in Oracle, leveraging the LLM to organize the schema and apply topological ordering to the SQL output. The LLM also generates datasets for testing the CFIHOS validator implementation. The results of this phase showed that the most significant issue in the generated SQL was the lack of correspondence between certain attributes with reference integrity. We then used the LLM to generate synthetic data, including code snippets for importing data into the RDL Excel file and SQL generation from scratch, ensuring compliance with referential integrity constraints.