Design and Evaluation of an AI-Driven Workflow for Technical Debt Remediation in Software Systems
Khalil, Rehan (2025-07-08)
Design and Evaluation of an AI-Driven Workflow for Technical Debt Remediation in Software Systems
(08.07.2025)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
avoin
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe2025073080111
https://urn.fi/URN:NBN:fi-fe2025073080111
Tiivistelmä
Technical debt (TD) is a persistent challenge in software engineering. It impacts long-term maintenance, reliability and security of the software as well as developer productivity. Many static analysis tools can identify debt, but taking actions, especially on the large scale, often requires a lot of work. This thesis explores how Large Language Model (LLM)-powered AI systems can automatically fix technical debt in real-world software.
This study investigates the application of AI-assisted tools in improving the maintainability of modern web applications. Using a case study of a progressive web app with identified technical debt, the project involved automated analysis using static code analyzer and security code scanner to detect code quality issues such as code smells, bugs and security vulnerabilities. The collected insights were then used in LLM-powered agentic AI to assess the effectiveness in remediating those issues.
Empirical results show that AI resolved approximately 90% issues, significantly improving the maintainability index of the project, by reducing code smells and bugs, and improving security vulnerabilities. However, the AI faced challenges in solving complexities requiring architectural decision and human validation. The study also show that it is possible to integrate diagnostic tools and AI agents into a continuous maintenance process.
This thesis offers a methodology for automated repayment of technical debt. It provides a critical assessment of AI tools capabilities in software maintenance and a better understanding of sustainable development practices with smart tooling. It concludes that while full autonomy is not yet possible, but LLM-powered tools are a promising step toward more effective and efficient technical debt management and repayment.
This study investigates the application of AI-assisted tools in improving the maintainability of modern web applications. Using a case study of a progressive web app with identified technical debt, the project involved automated analysis using static code analyzer and security code scanner to detect code quality issues such as code smells, bugs and security vulnerabilities. The collected insights were then used in LLM-powered agentic AI to assess the effectiveness in remediating those issues.
Empirical results show that AI resolved approximately 90% issues, significantly improving the maintainability index of the project, by reducing code smells and bugs, and improving security vulnerabilities. However, the AI faced challenges in solving complexities requiring architectural decision and human validation. The study also show that it is possible to integrate diagnostic tools and AI agents into a continuous maintenance process.
This thesis offers a methodology for automated repayment of technical debt. It provides a critical assessment of AI tools capabilities in software maintenance and a better understanding of sustainable development practices with smart tooling. It concludes that while full autonomy is not yet possible, but LLM-powered tools are a promising step toward more effective and efficient technical debt management and repayment.