AI-Assisted Software Repository Understanding and Intelligent Code Analysis applying Machine Learning and Large Language Models
Department of Artificial Intelligence and Data Science, Ajeenkya DY Patil School of Engineering, Pune, India · India
Abstract
Growing codebases force developers to spend more and more effort working out how the store is organised, to a significant degree, and also, where its weak points are. Current tooling tackles this one slice at a time, like rule-based analysers count metrics and flag violations, learned models rank modules by likelihood of defects, in real-world scenarios; also, large Language Models (LLMs) are outlining what code does in plain language. In this broader context, few of all these slices are connected, and most operate on one file at one more time; this review surveys ten recent papers on LLMs in software engineering, neural defect prediction, LLM-based judgement of maintainability, retrieval over code for repository-scale context, and agent-based defect prediction; checking them uncovers that no open, reproducible pipeline has been gauged that links repository reading, metrics, learned risk ranking, LLM explanation, and recommendations in one single output. A conceptual architecture is, in fact, therefore sketched in which a dependency graph and code metrics are driving a risk model; also, only the riskiest modules are then explained by an LLM applying setting drawn from that graph; this architecture is a proposal on paper. Nothing has been realized. Still tested yet, and a prototype and evaluation sketch are given for follow-up work.
Keywords
How to Cite This Article
Rights
BODHIVRUKSHA JOURNAL OF DIVERSE DISCIPLINE