Annotation

Annotation: Machine learning technologies and various tools for code generation have had a significant impact on the field of software development in recent years. Although most of the existing solutions are not built exactly for code generation, programmers apply them in different tasks. Not many of the existing AI solutions work well with less common languages, such as Kotlin or Swift, that are used in mobile development. Therefore, existing large language models are rarely adapted in the third-party software for mobile developers, although it would benefit the industry. The goal of this work is to develop a service that would use a large language model to provide the users, mobile developers, with a tool for efficient programming in the aforementioned programming languages. The developed service utilizes an already existing language model, which is fine-tuned based on the data available online in open-source repositories and collected manually. The developed software can perform various programming tasks specific to the mobile development domain: writing code for screen layouts, UI (User Interface) components, business logic, and unit tests. The software is also evaluated against the HumanEval benchmark and its variations as well as a custom benchmark that gives an understanding of the quality of generated code. This article is the result of a research project implemented within the framework of the fundamental research program of the National Research University Higher School of Economics (HSE University).