Životní cyklus znalostního grafu nad technickou dokumentací s využitím malých lokálních jazykových modelů

This diploma thesis presents the design, implementation, and evaluation of a knowledge graph lifecycle management system for technical documentation using small local language models. The main goal was to create a sustainable framework that overcomes the structural limitations of standard vector-based retrieval when querying hierarchically organized documentation, while maintaining full data sovereignty. The thesis combines theoretical knowledge of RAG architectures and knowledge graphs with the practical implementation of a hybrid system built on a graph database and a locally operated language model. As part of the work, an incremental graph management mechanism, a hybrid query pipeline, and an interactive user interface are designed and implemented. The contribution of the proposed architecture is validated through systematic evaluation using standardized metrics. The results may serve as a starting point for organizations considering the development of knowledge systems over structured documentation without dependency on commercial cloud models. The thesis contributes to understanding under what conditions and for what query types a graph-based approach yields measurable improvement over a standard vector baseline.