This paper introduces a method for constructing semantic networks in the field of cybersecurity using Large Language Models (LLM) based on the concept of a "swarm of virtual experts." The method involves multiple interactions with LLM through various prompts and roles, where each prompt is treated as a separate "virtual expert." This approach minimizes errors such as hallucinations and redundant links by aggregating results from different models and prompts. Additionally, a validation step involving human experts is introduced to enhance the reliability of the data obtained. The proposed approach demonstrates significant potential in the automated processing of complex data and can be applied to create semantic networks in cybersecurity tasks, data analysis, and information retrieval. Keywords: Swarm of Virtual Experts, Large Language Models (LLM), Cybersecurity, Semantic Networks, Automated Analysis, LLM Hallucinations, Data Validation