AUTOMATED GENERATION OF TECHNICAL DOCUMENTATION IN THE IT INDUSTRY USING LARGE LANGUAGE MODELS

Abstract

Automated generation of technical documentation is one of the key challenges in the modern IT industry, as high-quality and up-to-date documentation is essential for the effective work of developers, testers, and end users. This paper explores the potential of using large language models (LLMs) for generating technical documentation and evaluates the role of the ontological approach in structuring the documentation process.It has been noted that traditional methods of documentation creation face several difficulties, including a significant amount of manual labor, challenges in maintaining content relevance, and ensuring consistency across different sections. The lack of integration between documentation and software development processes leads to outdated materials and complicates their updates. Additionally, it is emphasized that in fast-paced development cycles, developers often pay insufficient attention to documentation, making it difficult to keep it up to date.Special attention is given to the conceptual integration of LLMs with ontological approaches, which allows not only the generation of texts but also ensures their structured organization according to formal knowledge models. The use of ontologies enhances documentation accuracy, reduces errors, and helps establish a unified documentation standard. This study presents an overview of approaches to integrating LLMs with ontologies, including automated knowledge extraction methods and the application of ontological modeling to documentation structuring.The research also analyzes the advantages and limitations of automated documentation generation. Key benefits include faster document creation, standardization, reduced workload for developers, and fewer human errors. However, several challenges are identified, such as the need for model training on domain-specific corpora, ensuring text accuracy, and integrating LLMs into the software development lifecycle.A separate section of the study is dedicated to analyzing practical cases of using LLMs for documentation generation. An example is presented of automatically creating documentation for an open-source software project written in Python, where GPT-4 was used to generate technical descriptions and user instructions. The results indicate that language models significantly improve documentation quality by ensuring a clear structure, proper formatting, and compliance with widely accepted standards. However, it is noted that for full compliance with official requirements, additional text editing and the inclusion of more detailed descriptions of parameters and variables may be necessary.The paper also discusses the future prospects of documentation automation, particularly through hybrid approaches that combine language models, ontologies, and quality control algorithms. It highlights the potential integration of LLMs into development environments (IDEs), which would allow automatic updates to documentation as code changes occur.It is emphasized that automating technical documentation generation using LLMs and ontologies is a promising direction that significantly increases developer efficiency, minimizes errors, and reduces the time required to maintain documentation. However, to ensure high accuracy and compliance with technical standards, further research is required in adapting language models to specific domains, developing methods for text accuracy control, and integrating automatic documentation systems with existing DevOps and code management tools.