Back to search

Institutional Repository of Vilnius University

Kodo kontekstualizavimo tobulinimas didžiųjų kalbos modelių padedamo programų sistemų kūrimo procese /

Abstract

dc:description

Nowadays, large language models (LLMs) are increasingly being used in the programming process to accelerate coding task completion. However, their limited understanding of the broader project context oſten results in generated code that, while syntactically correct, may be semantically incorrect or inconsistent with the existing project architecture and conventions. This study investigates the impact of providing structured additional information describing project architecture models, functional dependencies, and programming standards to LLMs in order to mitigate these issues. The study conducted an empirical investigation evaluating the latest generation of LLMs (”Anthropic” ”Claude Sonnet 3.7” and ”OpenAI” ”GPT­4o”) used in modern, AI­powered integrated development environments (”Cursor”, ”VS Code” with ”Github Copilot”) while performing typical soſtware development tasks. The activities were compared according to scenarios: in one, the LLM operated using standard contextual analysis capabilities provided by the agent mode developers, while in the other, it was supplemented with selected, structured project information. The main results show that the effectiveness of such additional context depends heavily on the specific LLM and agent configuration used. The more advanced model, when provided with additional contextual data, showed a significant improvement in task performance and compliance with best practices used in the project. Another LLM did not always show improved performance and sometimes even showed poorer results. This shows that different models process and use large amounts of structured information differently. The study concludes that providing strategic context for the selected project can be a valuable tool for improving the performance of capable LLM, but emphasizes that the ability of LLM to interpret and apply such information is critical to success. It also highlights the potential for automating the generation and maintenance of such contextual aids for practical application.

Degree

thesis:*
Grantor dc:publisher
Institutional Repository of Vilnius University
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Mažeikis, Gabrielius,

Rights

dc:rights
Statement dc:rights
  • info:eu-repo/semantics/openAccess
Language dc:language
lit

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
Vilnius University
Base URL
epublications.vu.lt/oai
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Mažeikis, Gabrielius,. Kodo kontekstualizavimo tobulinimas didžiųjų kalbos modelių padedamo programų sistemų kūrimo procese /. Institutional Repository of Vilnius University, 2025. https://repository.vu.lt/VU:ELABAETD239031679&prefLang=en_US