Self-Healing CI/CD Pipeline Using LLM and Google ADK Framework

Nukalapati, Vinay (2025) Self-Healing CI/CD Pipeline Using LLM and Google ADK Framework. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (5MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

Build failures in Continuous Integration/Continuous Deployment (CI/CD) pipeline is one of the greatest bottlenecks of modern software delivery where it wastes precious developer’s time and leads to delays in the release cycle. This research introduces the design, implementation and evaluation of an autonomous self-healing system using large language models (LLMs) and agentic AI to detect, diagnose and recover from these failures of the system without human intervention. The system has a hierarchical architecture of six specialised agents coordinated via the Agent Development Kit (ADK), which runs an iterative fix-verify-decide loop in AWS infrastructure.

Comprehensive evaluation across 60 diverse build failures has shown 73% automated remediation success rate, which has been validated as it has achieved 80% Mean Time to Recover (MTTR) success, which was 42 minutes with manual intervention and only 8.6 minutes with autonomous intervention. Comparative analysis with state of the art baseline research performance patterns reveals where effectiveness of the system is highly correlated with availability of error localization. Furthermore, results establish that frontier models (e.g., GPT-4) are critical in production readiness, much better than previous generations. These results prove for the first time that autonomous software engineering is not only an aspirational concept, but a working reality for well-scoped tasks, and it is a scalable solution for resolving the CI/CD reliability issues.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Lugones, Diego UNSPECIFIED
Uncontrolled Keywords:	Self-healing CI/CD; Agent Development Kit (ADK); LLM automation; pipeline orchestration; autonomous DevOps
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 09:26
Last Modified:	01 Sep 2026 09:26
URI:	https://norma.ncirl.ie/id/eprint/9723

Actions (login required)

View Item