Improving Telecom Solution Configuration using SysML v2 and GenAI

Sammanfattning

Telecom Solution Configurations (SC) face increasing complexity, leading to a lack of traceability and explainability in the current constraint-based configuration processes. This thesis investigates the potential of System Modeling Language v2 ( SysML v2) and Generative Artificial Intelligence ( GenAI), specifically large language model ( LLM)s, to improve SC. User-friendliness, interoperability, traceability and explainability are key aspects this thesis investigates. The research involved developing LLM-driven pipelines to: (1) dynamically represent telecom hardware data in SysML v2 from user prompts, (2) translate configuration requirements into SysML v2 and use LLM agents with tool-calling capabilities to explain constraint engine failures; and (3) generate hardware suggestions with reasoning based on SysML v2 defined requirements and hardware data. The suitability of SysML v2 for native constraint evaluation was also compared against Minizinc. Results demonstrate that the SysML v2 and LLM combination successfully created dynamic, user-friendly hardware data representations. The LLM pipeline provided accurate explanations for constraint failures in tested scenarios, enhancing traceability. LLMs also generated viable, and sometimes optimal, hardware suggestions with valuable reasoning, also enabling nuanced preferences when searching for hardware suggestions. However, native SysML v2 constraint evaluation proved significantly slower and more verbose than Minizinc for the tested constraints, deeming it currently infeasible for direct constraint solving. The study concludes that SysML v2, leveraged by GenAI , shows promising potential for improving telecom SC . While SysML v2 is not yet optimal for direct constraint execution, its synergy with LLMs can lower the domain knowledge threshold and increase the efficiency and transparency of the SC process.