Context-aware code review: integrating generative AI for automated pull request analysis

Abstract

Pull request reviews in software industry are vital for ensuring code quality. Traditional manual reviews offer valuable human insights but can be inefficient. They also struggle with the hallenges posed by rapidly growing, complex codebases. On the other hand, many automated tools focus only on syntax and style. They do not account for the broader business context. This paper presents a context-aware PR review system that combines generative AI, transformer-based embeddings, vector databases, and git diff augmentation to bridge the gap between technical accuracy and business needs. The goal is to provide clear feedback on both code implementation and intent, addressing challenges in large domain-specific codebases.