CodeQA: Advanced Programming Question-Answering Using LLM Agent and RAG

Abstract

In light of this complex information for programming environments, this paper explores how effectiveness in Large Language Models and concepts from Retrieval-Augmented Generation can be used to augment reduced hallucinated question-answering systems for programming environments. The present scenario of transformer-based models, though a big player in natural language processing, doesn't seem to have adaptability and context sensitivity, which is, in fact, a very important feature for a specialist domain. Our study, by integrating a sophisticated LLM model within the RAG framework, further improves the precision, effectiveness of retrieval, and sensitivity of the context of responses. The model based on LLM improves over this transformer-based model by scoring higher in accuracy and context awareness, supported by its pretraining on a massive corpus and dynamic document retrieval functionality. This result underlines the possibility of great improvement in QAS performance, handling very complex and specialized queries, through the integration of LLMs with the RAG systems, and thereby helps in indicating promising areas for further research in optimizing these methodologies across different domains.