Enhanced CodeT5 for the generation of Kubernetes configuration and actionable feedback

Ujjineni, Nagaravali (2025) Enhanced CodeT5 for the generation of Kubernetes configuration and actionable feedback. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (800kB) | Preview

Abstract

Although recent developments in the field of large language models (LLMs) and transformer-based code generators represent a significant advancement, they primarily generate syntactically correct settings but do not offer practical quality feedback or implement best practices in the domain. This paper fills this gap by developing a more refined CodeT5 system that is able to produce Kubernetes YAML specifications based on natural language specifications, in addition to measuring quality and giving detailed recommendations on how to make changes. The important innovations are a dual-encoder architecture, which offers enhanced semantic alignment between linguistic intent and Kubernetes configuration patterns, and an actionable feedback system that is based on eleven best practices in security. The enhanced model achieves 68.3% CodeBLEU, 100% YAML validity, and statistically significant improvements (p=0.020, Cohen’s d=1.13) over the baseline. The trade-off of the model between validity and quality is as follows: In cases of clear-intent best practices, compliance increased by a maximum of 63.6 percentage points, but in cases of unspecified intent, the model works on creating minimal configurations that are valid, as opposed to attempting to guess at what requirements have not been specified. These findings support the claim that specialised architectural improvements can significantly enhance the quality of automated Kubernetes configuration generation and also demonstrate significant design choices to make to address input uncertainty.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Estrada, Giovani UNSPECIFIED
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence T Technology > T Technology (General) > Information Technology > Cloud computing
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	01 Sep 2026 11:53
Last Modified:	01 Sep 2026 11:53
URI:	https://norma.ncirl.ie/id/eprint/9745

Actions (login required)

View Item