This work introduces a scalable, retrieval-augmented pipeline that integrates large language models (LLMs) with iterative refinement strategies to enable precise and efficient development of knowledge graphs (KGs). Leveraging domain-specific glossaries as seed input, an initial KG is generated and subsequently enhanced through GraphRAG-based enrichment. Validation is supported by automated frameworks that incorporate syntactic constraints and semantic similarity metrics to minimize human intervention while ensuring accuracy and reliability. The solution is evaluated using technical documentation from the National Cooperative for the Disposal of Radioactive Waste (NAGRA), a domain with stringent requirements for semantic integrity and factual precision. Results confirm that the proposed approach enables highquality KG construction suitable for semantic verification tasks, while also demonstrating adaptability across specialized domains and scalability in validation processes.