Development of an Intelligent Model Based on Generative Neural Networks for the Analysis of the Repository Code and Assistance to Programmers

In modern IT companies, sometimes situations arise when compromise decisions are made in order to develop functionality within the allotted time. Such solutions are often ineffective and lead to the accumulation of technical debt. The uncontrolled growth of technical debt entails architectural degradation of the project and an increase in the time spent on making changes. Reducing technical debt is a routine operation that requires significant time resources of the developer. This process is partially automated by the practice of continuous integration, but existing tools are effective only in a limited range of tasks. The current research focuses on the possibility of automating the process of reducing technical debt using large language models. In this paper, various large language models for correcting and generating program code are considered. In particular, a solution based on one of the considered neural networks will be proposed, including an application.