Annotation

Annotation: The article proposes an approach to the automated synthesis of the electronic structure of a product based on large language models (LLM), constraint programming and intermediate representation translators. The problem of converting unstructured natural language requirements of experts into formalized machine-readable representations conforming to GOST R 2.053-2023 and GOST 2.054-2013 standards is considered. A key feature of the approach is the use of LLM as a semantic converter that generates an intermediate representationin accordance with a formal context-free grammar based on JSON. The architecture of the system is presented, which includes modules for semantic synthesis, validation of intermediate representation and translators into target formats. The paper presents a modular system architecture that includes three main levels: interaction, processing and synthesis, as well as data and integration. The components of synthesis, validation of intermediate representation, and translators into SQL and UML formats are described in detail. Two algorithms have been developed and experimentally tested: the basic synthesis of ESI and an extended one using the human-in-the-loop methodology for iterative refinement of requirements. The practical significance of the work lies in the possibility of integrating the developed algorithms into automation systems for design, technological preparation of production and product lifecycle management in industrial enterprises.