DEVELOPMENT OF LLM-BASED ASSISTANCE TOOL FORCODE REVIEW

• Main Author: Khoiron Fadhila, Fikri
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85040
• Institution: Institut Teknologi Bandung

In software development, the code review process is critical for maintaining quality and security. This research identifies key challenges in this process, including difficulties in understanding the reasons behind code changes and limitations of existing review tools. Therefore, this thesis proposes the use of LLMs in developing a tool to enhance the efficiency and effectiveness of code reviews by providing deeper insights into code changes and automatically generating context-aware improvement suggestions. This approach helps developers understand the reasons behind changes and accelerates the review process. The methodology involves several stages: requirements analysis, design, integration with LLM, and evaluation. The tool's design is modeled using C4 to ensure structural clarity. The tool is implemented as a web-based application integrated with an LLM. The LLM is trained using prompting techniques and prompt optimization through prompt engineering. The tool is tested through functional testing across various scenarios. The test results show that the tool performs well in understanding code changes and providing context-aware improvement suggestions. Thus, this LLM-based code review assistance tool offers a potential solution to support the code review process, making software development more efficient and effective.