Code generation and repo understanding using LLMs, knowledge graphs and search-based algorithms

Abstract

The increasing complexity of codebases demands advanced methods for automated code gener ation and repository understanding. This project explores the synergistic application of large language models (LLMs), knowledge graphs, and search-based algorithms to address these chal lenges. A comprehensive literature survey was conducted, culminating in a published survey paper that highlights state-of-the-art methodologies in the domain. To facilitate experimenta tion, a novel dataset was curated by integrating four prominent datasets, tailored for fine-tuning the CodeT5 and QwenCoder models. These fine-tuned models aim to enhance code synthesis and repository analysis, offering improved accuracy and contextual understanding. This re search contributes a holistic approach to bridging gaps in automated code intelligence through innovative integrations of LLMs and auxiliary technologies.