Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

AI Powered Code Review Assistant

Alok Anand1; Anish Kumar2; Akash Babu NJ3; Deepak Kumar4

Internal Guide: Varshitha M K5

Assistant Professor, Department of ISE,
RNSIT Department of Information Science
RNS Institute of Technology, Bengaluru

Abstract:- With the increasing complexity of software code- bases. As software systems become more intricate and
systems, maintaining high code quality is essential to development timelines shrink, the limitations of manual code
ensure reliability, maintainability, and security. review processes have led to an increased interest in
Traditionally, code reviews havebeen a manual and time- automated solutions.
consuming process, often resulting in inconsistencies and
missed issues due to human error. Recent advance- ments Recent advancements in artificial intelligence (AI),
in artificial intelligence, specifically generative AI models particularly in natural language processing (NLP) and large
like OpenAI’s Chat- GPT and Google Gemini, have language models (LLMs), have introduced new possibilities
opened new possibilities for automating code reviews by for code review automation. Generative AI models like
providing real-time, intelligent feedback on code quality. OpenAI’s ChatGPT and Google’s Gemini API are capable of
understanding code struc- ture and context, making them
This survey paper explores the current state of AI- suitable for tasks such as code analysis, bug detection, and
assisted code review tools, focusing on the potential of suggesting optimizations. These models offer real-time
generative AI models to improve software development feedback within the developer’s integrated development
workflows. We exam- ine the methodologies, benefits, and envi- ronment (IDE), reducing the time and effort required for
limita- tions of existing tools such as GitHub Copilot, traditional code reviews. By leveraging these AI- powered
Amazon CodeWhisperer, and other AI-drivensolutions. tools, developers can benefit from automatedinsights on code
Additionally, we discuss the archi- tecture and design of quality and best practices without the need for constant
an AI-powered code review assistant that integrates human oversight.
seamlessly with popular development environments like
VS Code, leveraging cloud-based processing through This survey aims to explore the potential and currentstate
AWS. of AI-driven code review systems, focusing on generative AI
models as tools to improve developer productivity and code
Our findings suggest that integrating gener- ative AI quality. We examine popular tools like GitHub Copilot and
into the code review process can significantly reduce Amazon CodeWhis- perer, analyzing their methodologies,
review time, improve consistency, and enhance developer advantages, and limitations. Additionally, we propose an
produc- tivity. This paper also highlights the cost-effective architectural design for an AI-powered code review assistant
implementation of AI models in code reviews, in- tegrated with VS Code, leveraging cloud-based re-sources
demonstrating the feasibility of deploying scalable, (e.g., AWS) for scalable processing and real- time feedback.
budget-friendly solu- tions in real-world applications. By
analyz- ing the strengths and weaknesses of current In summary, this paper investigates the impact of AIon
approaches, we outline the path for futureadvancements traditional code review practices and assesses the feasibility
in AI-powered code review systems, focusing on multi- of adopting generative AI models in real- world development
language support, enhanced security analysis, and environments. Through this study, we identify strengths,
continuouslearning capabilities. challenges, and areas for future research, including enhanced
multi-language support, improved security analysis, and
Keywords:- AI, Code Review, Generative AI, ChatGPT, adaptability to variouscoding standards.
GeminiAPI, Software Development.
II. LITERATURE SURVEY
I. INTRODUCTION
 In recent years, artificial intelligence (AI) and ma- chine
In modern software development, maintaining high learning (ML) have significantly transformed thesoftware
code quality is essential for ensuring the reliability, development process, particularly in the area of code
maintainability, and security of applications. Tradi-tionally, generation and review. As software systems become
code reviews have been conducted manu-ally by experienced increasingly complex, ensuring code quality and
developers who inspect code for bugs, style adherence, and maintaining development efficiency are critical
security vulnerabilities. While effective, manual code challenges for developers and organizations alike. AI-
reviews are often time- consuming, prone to inconsistencies, driven code review tools aim to automate parts of the code
and susceptible to human error, especially in large, complex review process, providing developers with real- time

IJISRT24DEC156 www.ijisrt.com 458

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
feedback, improving code consistency, and en- hancing  Flask Documentation: Flask is a lightweight web
productivity. This literature survey examines various framework for Python, often used to build APIs and
studies on AI-assisted code review, analyzing their backend systems [6]. The official Flask documentation
methodologies, benefits, and limitations. Through these serves as a valuable resource for developers building
studies, we gain insights into the current state of AI in backend servers for code review assistants, allowing
software development and the potential of tools like smooth integration with AI models hosted on cloud
generative AI models to advance code qualityassurance. platforms like AWS. Its simplicity and flexibility make
 Evaluating the Code Quality of AI-Assisted Code Flask an ideal choice for rapid development.
Generation Tools: Coutinho et al. [1] conducted an  GitHub Documentation: GitHub provides de- tailed API
empirical study evaluating code generation tools such as and developer tool documentation that supports the
GitHub Copilot, Amazon CodeWhisperer, andChatGPT. integration of AI code review tools withina development
The study assessed these tools based on their impact on workflow [7]. GitHub’s documentationincludes guidance
code quality, focusing on common errors and style on using its REST API, which enables automation of pull
adherence. The findings highlight that AI-driven tools request reviews and feed-back generation. This is
can significantly enhance coding efficiency, but their particularly beneficial for AI- assisted code reviews,
reliability varies across different coding tasks. A primary allowing real-time collaboration and continuous
limitation is the lack of nuanced understanding in some integration.
complex scenarios, where human judgment remains  ChatGPT API Documentation: OpenAI’s Chat-GPT API
essential. documentation offers insights on integrat- ing ChatGPT
 From Google Gemini to OpenAI Q: A Survey on into applications for various purposes, including code
Generative AI: Han et al. [2] provide an exten- sive survey generation and analysis [8]. The API supports natural
on the advancements in generative AI, including models language interactions with code, mak- ing it a valuable
like Google Gemini and OpenAI’s Q*. This survey asset for building AI-driven code review assistants.
highlights the applicability of these models in various However, the documentation also notes limitations on
domains, including code analysis and generation. One response lengths and rate limits, which may impact its
advantage is their ability to generate context-aware effectiveness in large-scale applications.
suggestions, which is valuablefor developers. However,  Google Gemini API Documentation: The Google Gemini
the study also notes that real- world implementation API documentation provides infor- mation on using
challenges persist, particularly in adapting generative Gemini models for tasks such as code analysis and
models for industry-specificrequirements. feedback generation [9]. Gemini offers a powerful
 A Comparative Review of AI Techniques for Automated generative AI model that can be customized for various
Code Generation: Lee et al. [3] offer a comparative domains, including software development. The primary
analysis of machine learning and natural language challenge, as mentioned in the documentation, lies in
processing techniques used in code gener- ation. This setting up adequate securityand ensuring compliance with
paper reviews foundational approaches,such as rule-based usage policies whendeploying Gemini-based solutions.
systems and neural network-based models, and their  In summary, the literature suggests that while AI- driven
evolution to modern AI-driven tools. The review code review tools are rapidly advancing, they are not
emphasizes the effectiveness of AI in automating without limitations. Key challenges include adapting
repetitive tasks, yet acknowledges limita- tions in these tools to handle complex coding scenar-ios, ensuring
handling complex code structures that require contextual multi-language support, and maintainingthe flexibility to
understanding. integrate with various development environments. These
 The Impact of Artificial Intelligence on Pro- grammer insights form the basis for devel-oping a more robust and
Productivity: Kumar and Chen [4] explorethe impact of adaptable AI-powered code review assistant, leveraging
AI tools on developer productivity, with a focus on both generative AI modelsand scalable backend services.
automating repetitive coding tasks. Their survey
demonstrates that AI-driven tools can streamline III. OBJECTIVES
workflows, reduce time spent on manual reviews, and
improve productivity. However, the studypoints out that The primary objective of this project is to develop an
such tools are currently limited in sup- porting certain AI-powered code review assistant that enhances software
programming languages and complex codebases, which development workflows by providing real- time, contextual
affects their universal applicability. feedback to developers. The systemis designed to automate
 AWS EC2 Documentation: The official doc- umentation repetitive tasks, improve code quality, and reduce the time
for Amazon Web Services (AWS) EC2 provides spent on manual code reviews. The key objectives of this
comprehensive guidance on deploying cloud- based project are as follows:
resources [5]. For AI-powered code review systems, the
AWS EC2 free tier, specifically the t2.micro instance,  Automate Code Review Process: Utilize gener- ative AI
enables affordable hosting solutions for small-scale models to streamline the code review process, enabling
applications. The resource outlines setup processes and developers to receive instant feedback on code quality,
best practices, offering scalabilityfor projects needing on- structure, and potential bugs.
demand computational power.

IJISRT24DEC156 www.ijisrt.com 459

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

 Improve Developer Productivity: Minimize the time The assistant operates by capturing code snippets within
developers spend on manual reviews, allow- ing them to the development environment, sending themto the backend
focus on complex tasks and enhancing overall for analysis, and returning actionable feedback to the
productivity. developer. This setup minimizes man- ual intervention,
 Enhance Code Quality and Consistency: Ensurethat AI- enhancing workflow efficiency while maintaining high
driven suggestions align with coding best practices, standards of code quality and con- sistency. Additionally, the
making code more readable, maintain- able, and secure. system’s design prioritizes affordability, using free or low-
 Seamless Integration with Development En- vironments: cost cloud resources tomake it accessible to individuals and
Implement the assistant as an ex-tension for popular small teams.
code editors like VS Code, ensuring a smooth and non-
intrusive experience within the existing development V. ADVANTAGES OF PROPOSED SYSTEM
workflow.
 Cost-Effective Implementation: Demonstrate that the AI- The proposed AI-powered code review assistant
powered solution can be developed and deployed within provides several advantages over traditional code re- view
budget constraints using re- sources like AWS Free Tier, methods. These benefits make it an effective, efficient, and
making it accessible for individuals and small teams. accessible solution for developers. The key advantages are
as follows:
These objectives aim to demonstrate the feasibility and
effectiveness of integrating AI into the code review process,  Real-Time Feedback: Provides developers with instant
providing a practical, scalable solutionfor modern software feedback on code quality, helping them address issues
development. immediately rather than waiting formanual review cycles.
 Improved Productivity: Reduces time spent on manual
IV. PROPOSED SYSTEM reviews, allowing developers to focus on more complex
tasks and accelerating the develop-ment process.
The proposed system is an AI-powered code review  Enhanced Code Quality: Uses AI-driven insights to
assistant that integrates with popular development enforce coding standards and best practices, improving
environments, such as VS Code, to provide real-time, code readability, maintainability, and security.
automated code feedback. Leveraging generative AI models  Scalability: The cloud-based backend hosted on AWS
like ChatGPT or Google’s Gemini API, the system aims to enables scalable code review processes, making it suitable
identify bugs, optimize code struc-ture, and enforce coding for both individual developers and large teams.
best practices. This system utilizes a cloud-based backend,  Cost-Effective Solution: By leveraging free-tier cloud
hosted on AWS, to handle code analysis requests, ensuring resources and usage-based AI models, the system remains
scalability andconsistent performance. affordable and accessible, espe- cially for small teams
and individual developers.
 Seamless Integration with IDEs: Embeds di- rectly into
development environments like VS Code, providing a
smooth and uninterrupted workflow for developers.

Fig 1: Comparison of Time Spent on Manual vs. AI-Assisted Code Reviews

IJISRT24DEC156 www.ijisrt.com 460

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
VI. PROPOSED METHODOLOGY VII. SYSTEM DESIGN

 The Methodology for Developing the AI-Powered Code The system design of the AI-powered code review
Review Assistant Consists of the Following Steps: assistant involves several components, including the VS
Code extension, AWS backend, and an AI model (e.g.,
 System Integration with VS Code: Develop an ChatGPT or Gemini) for code analysis. The architecture
extension for Visual Studio Code that allows developers facilitates seamless communication and real-time feedback,
to initiate code reviews directly within the IDE. The enhancing developer productivity and code quality.
extension will capture code snippetsor files and send them
to the backend for process-ing. A. System Requirements
 Backend Processing and API Communication: Implement The system requirements are divided into client-side(VS
a backend server using Flask, hosted on AWS EC2 Code) and server-side (AWS backend) specifica- tions.
(t2.micro instance). This backend will handle requests
from the VS Code extension and communicate with the  Client-Side (VS Code Extension):
AI model API (suchas ChatGPT or Gemini) for code
analysis andfeedback generation.  Visual Studio Code editor with support for exten-sions.
 AI Model Analysis and Feedback Generation: Leverage a  Internet connection to communicate with backend
generative AI model to analyze the received code for services.
bugs, style improvements, and security issues. The  VS Code extension with UI to trigger code review
model’s output will include recommendations and requests.
improvements, which are then formatted by the backend
for clarity.  Server-Side (AWS EC2 Backend):
 Feedback Display and User Interaction: The processed  t2.micro instance on AWS EC2 (Free Tier eligi-ble).
feedback is returned to the VS Code extension, where it is  Flask framework to handle API requests.
displayed in an easy-to-understand format. The developer  Python environment with libraries for AI model
can review andapply the suggested changes, streamlining integration.
the code review process with minimal manual inter-
vention. B. System Architecture Diagram
The architecture of the AI-powered code reviewsystem
This methodology ensures a seamless integration of the is depicted in Figure VII.2, illustrating the interactions
code review assistant into the developer’s work- flow, between the Developer, Code Review Sys- tem, and AI
providing real-time, actionable feedback and enhancing Model.
overall productivity.

Fig 2: System Architecture for AI-powered Code Review Assistant

IJISRT24DEC156 www.ijisrt.com 461

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
VIII. DATA FLOW DIAGRAMS (DFDS) IX. CONCLUSION

A. Context Diagram (Level 0 DFD) The AI-powered code review assistant demonstratesthe
potential of integrating artificial intelligence into software
 Purpose: development workflows, specifically in au- tomating and
Displays the system at a high level, illus- trating the enhancing code review processes. The project achieves
interactions between the main components (VS Code, AWS several goals:
backend, and AI model).
 Reduces manual code review time, enabling de- velopers
 Components: to focus on complex tasks.
 Improves code quality and consistency through real-time
 User/Developer: Initiates the code review. feedback based on AI analysis.
 VS Code Extension: Receives and forwards code  Seamlessly integrates into popular development
snippets. environments like VS Code.
 Backend System (AWS EC2): Processes the codewith AI  Demonstrates the feasibility of deploying AI- driven
assistance. solutions within budget using cloud ser- vices.
 AI Model: Provides analysis and returns feedbackto the  Provides a foundation for future improvements in
developer. automated code quality assurance.
 Overall, this project showcases how AI can be a valuable
B. Level 1 DFD - Expanded View tool in modern software engineering, sup- porting
developers with timely, contextual feedbackto improve
 Expanded View: productivity and maintain high coding standards.

FUTURE ENHANCEMENTS

The current system provides a strong foundation,but

there are several areas for future improvements and

 Data Flow: Shows each stage of data processingand additional features that could enhance its utility:
how it is transformed at each step
 Multi-Language Support: Extend support to more
programming languages for broader appli- cability.
 Enhanced Security Analysis: Incorporate secu- rity
checks to detect potential vulnerabilities.
 Project-Level Analysis: Enable the assistant to analyze an
entire project directory, not just indi- vidual files.
 Continuous Learning: Integrate machine learn- ing to
improve suggestions based on user feed- back and coding
patterns.
 Improved User Interface: Develop an interactiveUI with
more detailed explanations and guidance on suggested
changes.

These enhancements would further the AI assistant’s

capabilities, making it an even more comprehensive and
adaptive tool for developers in various domains.

REFERENCES

[1]. M. Coutinho, L. Marquez, and F. Wang, “Evaluating

the Code Quality of AI-AssistedCode Generation
Tools: An Empirical Studyon GitHub Copilot,
Amazon CodeWhisperer, and ChatGPT,” arXiv,
2023. Available:
https://ar5iv.labs.arxiv.org/html/2312.10868
[2]. T. R. McIntosh, T. Susanto, and L. Brown, ”From
Fig 3: Level 1 DFD - Expanded View of the Google Gemini to OpenAI Q: A Survey on Gen-
Code Review Process erative AI,” 2023. Available: https://alok180202-
my.sharepoint.com

IJISRT24DEC156 www.ijisrt.com 462

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[3]. Odeh, N. Odeh, and R. Ahmed, ”A Compara- tive
Review of AI Techniques for Automated Code
Generation,” Journal of Recent AI Advancements,
vol. 13, no. 1, pp. 726-739, 2023.
[4]. R. Ferdiana, ”The Impact of Artificial Intelligenceon
Programmer Productivity,” ResearchGate, 2024.
Available: https://www.researchgate.net
[5]. Amazon Web Services, ”Amazon EC2 User Guidefor
Linux Instances,” AWS Documentation. Avail-able:
https://docs.aws.amazon.com
[6]. Pallets Projects,”Flask API Documen-tation,” Flask
Documentation. Available:
https://flask.palletsprojects.com/en/stable/api/
[7]. GitHub,Inc., ”GitHub API Docu- mentation,”
GitHub Docs. Available:
https://docs.github.com/en/rest
[8]. OpenAI,”ChatGPT API Documenta- tion,” OpenAI
Documentation. Available:
https://platform.openai.com/docs/overview
[9]. Google,”Google Gemini API Documen- tation,”
Google Documentation. Available:
https://cloud.google.com/gemini/docs
[10]. Microsoft, ”VS Code Docu- mentation,” Microsoft.
Available:https://code.visualstudio.com/docs

IJISRT24DEC156 www.ijisrt.com 463