Papers

Prescriptive and Contextual Technical Debt Management with LLM and SonarQube

Technical Debt (TD) is a short-term technical decision that compromises long-term software quality and stability. Static analysis tools can identify TD but only provide diagnostic analysis without prescriptive recommendations or project-specific context. This research develops an LLM-based TD management system integrated with SonarQube and GitHub to provide prescriptive and contextual analysis at the pull request level. The system leverages GPT-4 to analyze SonarQube issues within the context of pull requests, highlighting severity, short and long-term effects, and suggested remediation actions. Evaluation using a Rule2Text LLM-as-a-judge framework on 108 TD issues from two Apache projects achieved 96% good/very good scores for prescriptive analysis and 64% for contextual analysis.