Integrace velkých jazykových modelů do systému Kelvin

Abstract

This thesis focuses on the design and implementation of a module for automatic analysis of source code using large language models (LLMs) within a web-based information system used in programming education. The thesis describes the principles of LLMs, the Transformer architecture, and approaches for applying these models to source code analysis. It also compares different deployment approaches and prompting strategies with respect to output quality, operational requirements, and data protection. The proposed solution is experimentally evaluated on real-world data, demonstrating the potential of LLMs as a supporting tool for assessing programming assignments.