Exportar Publicação

Kumar, R. & Coutinho, C. (2026). Code Review with Large Language Models. In Proceedings of the International Conference on Electrical and Computer Engineering Researches (ICECER2025).

R. Kumar and C. E. Coutinho,  "Code Review with Large Language Models", in Proc. of the Int. Conf. on Electrical and Computer Engineering Researches (ICECER2025), 2026

@inproceedings{kumar2026_1773824980090,
	author = "Kumar, R. and Coutinho, C.",
	title = "Code Review with Large Language Models",
	booktitle = "Proceedings of the International Conference on Electrical and Computer Engineering Researches (ICECER2025)",
	year = "2026",
	editor = "",
	volume = "",
	number = "",
	series = "",
	doi = "10.1109/ICECER65523.2025.11401087",
	publisher = "",
	address = "",
	organization = "",
	url = "https://www.icecer.com/2025/"
}

TY  - CPAPER
TI  - Code Review with Large Language Models
T2  - Proceedings of the International Conference on Electrical and Computer Engineering Researches (ICECER2025)
AU  - Kumar, R.
AU  - Coutinho, C.
PY  - 2026
DO  - 10.1109/ICECER65523.2025.11401087
UR  - https://www.icecer.com/2025/
AB  - With the recent rise of conversational AI
(Artificial Intelligence) models, such as ChatGPT, it is essential
to understand how these models can be used to complete tasks
faster and more efficiently. Large Language Models (LLMs) can
assist software developers to solve a variety of problems such as
completing missing portions of code, finding vulnerabilities in
the code, styling the code. Therefore, these tools can be useful to
accelerate the learning curve. This paper studies how LLMs can
be used for code review by junior software developers. A code
review aims to identify mistakes that a junior developer may
pass and improve code’s readability. In this study a tool was
built that uses the ChatGPT API to review code. The same pieces
of code were reviewed using different prompts and different
versions of ChatGPT model. The results indicate that, to use
LLMs effectively for code review, the prompt alone is not
enough, there should also be used the model with the higher
number of parameters to obtain better reviews.
ER  -