This master’s thesis addresses the inefficiency of manual dashboard creation for digital twins (DT) of telecommunications network management systems (TNMS). It develops and evaluates an AI-driven pipeline that uses a large language model (LLM) augmented by a domain-specific knowledge graph (KG) and retrieval mechanisms (RAG) to automate the generation of dashboards in a compatible format.

Following the Design Science Research (DSR) methodology, the Awareness Phase provides dashboard requirements via literature and expert interviews. The Suggestion Phase introduces a model of a three-stage pipeline that turns raw network-management data into live managerial dashboards. The Development Phase builds upon the previous phase by creating a functional prototype. The prototype processes real network data, formulates precise queries, and produces ready-to-use dashboard definitions. The Evaluation Phase focuses on assessing the prototype through practical scenarios and expert reviews. The assessments and interview feedback indicate that time savings in dashboard production, reliable handling of various query requests, and improved consistency across visualizations were clear outcomes. Challenges include handling unclear input and securing high-reliability domain data. Limitations during the final assessment of the prototype were the number and selection of interview experts. The user feedback was collected exclusively from the engineering department, excluding the operations teams whose perspectives might have differed in specific aspects. Another limitation is the artifact’s reliance on pre-existing network-management ontologies. The reliance means, that any omissions or errors within those ontologies directly affect the fidelity of generated dashboards.

The final results demonstrate the viability of an LLM KG-RAG approach for automating dashboard creation in network DT and contribute a replicable model for practitioners and researchers aiming to integrate AI into network management workflows.