The development of large language models (LLMs) has rapidly transformed the extent and ease with which information extraction can be automated in many domains. Built on enormous datasets, they can identify patterns in textual data. However, many industrial domains rely on limited and specialized expert data and documentation that are not easily accessible with LLMs alone. Traditionally, this type of data has been approached with knowledge representation methods and semantic web technologies. These methods are labor-intensive as they require human experts to manually create and curate the content. One emerging area of research is to use the efficiency of LLMs to aid in the construction of semantic information and knowledge graphs (KGs). However, as LLMs are prone to hallucinations, the main challenge is to provide automated methods for knowledge generation that also verify the results with respect to original content and real-world commonsense. Addressing this challenge, we introduce MAVer-KG, a novel system for extracting KGs from natural language text using LLMs. The system is designed as a multi-agentic system in which LLMs extract information, text embeddings act as a commonsense verification filter, and an inductive reasoner agent allows for intelligently expanding the knowledge graph with novel, domain-relevant insights. In their machine-readable format, the extracted knowledge graphs can seamlessly be integrated into a range of application scenarios and/or be further manipulated. The system is demonstrated with a running example of extracting information from formal documents for automotive system engineering and Autonomous Driving System (ADS).