Artificial intelligence-supported code review system
An AI-supported system provides real-time, context-aware feedback and performance metrics for developers, addressing inefficiencies in pull request reviews by integrating with popular code platforms, thus optimizing software development processes.
Patent Information
- Application Number
- PCT/TR2026/050495
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-05-06
- Publication Date
- 2026-11-12
AI Technical Summary
Existing pull request review systems in open-source software development struggle to provide real-time, context-aware feedback and performance metrics for developers, leading to inefficiencies and increased workload for project managers.
An AI-supported system that analyzes code changes in real-time, providing line-by-line feedback and performance metrics by integrating with platforms like GitHub, GitLab, and Bitbucket, using a multi-layered AI architecture to evaluate code quality and developer contributions.
Enhances developer performance monitoring and reduces errors by offering immediate, context-specific feedback and metrics, optimizing the review process and reducing managerial workload.
Smart Images

Figure TR2026050495_12112026_PF_FP_ABST
Abstract
Description
[0001] DESCRIPTION
[0002] ARTIFICIAL INTELLIGENCE-SUPPORTED CODE REVIEW SYSTEM Technical Field
[0003] The present invention relates to a system for generating contextual code annotations and developer performance metrics by analyzing, in real time, code change requests generated during software development processes.
[0004] Background of the Invention
[0005] The pull request model, widely used in open-source software development processes today, enables project contributors to submit their independent code changes to the main repository. While this model encourages software collaboration, the increasing volume of pull requests poses a significant review and decision-making burden on project integrators. Some prediction models developed with the aim of alleviating this situation attempt to determine whether a pull request should be merged or not; however, these models typically rely on advanced data that can only be obtained after the request has been created. The inability to perform decision predictions based solely on existing basic features when a pull request is first created prevents the optimization of review processes and results in shortcomings in this area.
[0006] For this reason, in order to overcome the above-mentioned shortcomings, there is a need for a system which evaluates code changes in context and enables individual performance to be measured over time by providing real-time and instructive feedback to the developer.
[0007] The Chinese patent document no. CN117055856A, an application included in the state of the art, discloses a method for predicting the decisions on code changerequests at an early stage. The method in the said invention enables the prediction of the final decision, that is, acceptance or rejection, regarding code change requests (pull requests) made to a software project while the requests are still in the generation phase. The method is based on two main feature sets: the title and description text of the code change request, along with deep semantic features obtained from code changes, and manually extracted early features of the developer, the change request, and the project. Deep features are extracted as multidimensional vectors from description text and code changes, respectively, by means of pretrained models such as RoBERTa and CodeBERT. Manual features, on the other hand, consist of past activities of the developer, structural characteristics of the request, and general statistics of the project. The semantic representation is created by combining the obtained deep vectors with fully connected layers, and the decision prediction is made by sending them to the classifier along with manual features. The model is optimized by fine-tuning with labeled data during the training phase. Through this method, the workload of project managers in open-source projects is reduced, while contributors receive early feedback, thereby enhancing the efficiency of the collaborative software development process.
[0008] Summary of the Invention
[0009] An object of the present invention is to realize a system which encourages individual development and provides reporting to managers by measuring the contribution rates of developers, their impact on code quality, and the frequency of receiving feedback as metrics, by providing line-by-line feedback to developers by analyzing, in context, the code changes in pull requests created during software development processes.
[0010] Another object of the present invention is to realize a system which provides a scalable and platform-independent analysis infrastructure that can be integrated into the various development environments used by software teams, by also working ondifferent code management and version control platforms in the form of GitHub, GitLab, Azure DevOps, and Bitbucket.
[0011] Detailed Description of the Invention
[0012] “Artificial Intelligence-Supported Code Review System” realized to fulfil the objectives of the present invention is shown in the figure attached, in which:
[0013] Figure l is a schematic view of the inventive system.
[0014] The components illustrated in the figure are individually numbered, where the numbers refer to the following:
[0015] 1. System
[0016] 2. Database
[0017] 3. Server
[0018] A. Code management platform
[0019] A system (1) for generating contextual code annotations and developer performance metrics by analyzing, in real time, code change requests generated during software development processes comprises
[0020] at least one database (2) which is configured to keep a record therein of code changes made by software developers, artificial intelligence comments on these changes, the categorical classifications of the comments, developer contribution rates, and data related to the history of previous feedback; and at least one server (3) which is in communication and exchanges data with the database (2); runs at least one software thereon that automatically processes data and generates meaningful results; detects code differences and the methods to which they belong by retrieving pull request data from the code management platform (A) through the software running thereon; performs the interpretation process on these differences through a multi-layered, chained artificial intelligence architecture; and is configured to provide feedback on the analysis results to the authorized units by generating developer performance metrics.
[0021] The database (2) included in the inventive system (1) is configured to keep a record of code changes made by software developers, the artificial intelligence comments on these changes, the category to which each comment belongs, the contribution rate of the software developer, the history of received feedback, and metrics regarding the impact on quality in accordance with the pull request data received through the code management platform (A), by establishing communication with this platform that enables the code changes made by software developers to be monitored and to be managed through pull requests.
[0022] The server (3) included in the inventive system (1) is configured to run at least one artificial intelligence software thereon that automatically processes data and generates meaningful results. The server (3) is configured to establish communication with the database (2) and to record data in the database (2). In the preferred embodiment of the invention, the server (3) is configured to receive HTTP (HyperText Transfer Protocol) requests triggered by the processes of pull request creation and pull request updating performed by software developers on the code management platform (A). The server (3) is configured to automatically receive and process data related to pull request events via a developer notification connection tool (service hook structure), enabling these HTTP requests to be transmitted to the developer interface. The server (3) is configured to analyze code changes with metadata related to the source and target branch names of the pull request received with HTTP requests, the names of the modified files, which lines in each file have changed, and whether the file is new or an existing one. The server (3) is configured to establish the correct analysis context by determining the start and end boundaries of code blocks by applying syntax analysis to preserve the structural integrity of the code and identify the method to which each change belongs, by processing these specified line-based code differences. The server (3) is configured to evaluate, byusing an artificial intelligence model running on the code management platform (A) to analyze the obtained method-based code differences, the interpretation process of this model through a multi-layered artificial intelligence architecture that manages this process in layers and provides quality control at every stage. The server (3) is configured to enable the technical quality of the code to be evaluated based on security, readability, and performance criteria within the scope of multilayered artificial intelligence architecture, in order to step-by-step process and filter the comments obtained from the generative artificial intelligence model. The server (3) is configured to filter out ambiguous, unsupported, or unnecessary expressions that may appear in comments obtained from the generative artificial intelligence model. The server (3) is configured to check that the generated comments comply with current software standards and customer-specific rules. The server (3) is configured to enable only technically meaningful and truly necessary comments to remain in the system. The server (3) is configured to provide the developer with the overall context of the code changes made and to summarize all changes in an understandable manner. The server (3) is configured to generate comments that have been definitively validated and contextually appropriate by transferring the generated output to the next evaluation step, and to both transmit these comments directly to the relevant line of code through the code management platform (A) and to record them in the database (2). The server (3) is configured to record comments generated by the generative artificial intelligence model and stored in the system after their accuracy has been verified, along with the category information to which they belong, the identity of the developer the comment pertains to, the contribution rate, the frequency of receiving comments, historical feedback structures, and performance metrics reflecting their impact on code quality, into the database (2). The server (3) is configured to analyze the development process of the developer by comparing current data with previously recorded comment content, contribution rates, method-based quality analyses, and time-series data in the database (2), and to provide team managers with meaningful outputs on development trends.
[0023] Industrial Application of the InventionBy means of the inventive system (1), it is enabled to provide developers with realtime technical feedback by analyzing code changes made during software development processes; to improve quality control processes in software projects by measuring contribution levels and reporting this data to managers; to reduce error rates and to facilitate the monitoring of in-team performance, as well as the fact that it is designed with an architecture capable of being integrated with various version control and code collaboration platforms in the form of GitHub, GitLab, and Bitbucket, enabling it to be applied independently of the platform.
[0024] Within these basic concepts; it is possible to develop various embodiments of the inventive “Artificial Intelligence-Supported Code Review System (1)”; the invention cannot be limited to examples disclosed herein and it is essentially according to claims.
Claims
CLAIMS1. A system (1) for generating contextual code annotations and developer performance metrics by analyzing, in real time, code change requests generated during software development processes; comprisingat least one database (2) which is configured to keep a record therein of code changes made by software developers, artificial intelligence comments on these changes, the categorical classifications of the comments, developer contribution rates, and data related to the history of previous feedback; and characterized byat least one server (3) which is in communication and exchanges data with the database (2); runs at least one software thereon that automatically processes data and generates meaningful results; detects code differences and the methods to which they belong by retrieving pull request data from the code management platform (A) through the software running thereon; performs the interpretation process on these differences through a multilayered, chained artificial intelligence architecture; and is configured to provide feedback on the analysis results to the authorized units by generating developer performance metrics.
2. A system (1) according to Claim 1; characterized by the database (2) which is configured to keep a record of code changes made by software developers, the artificial intelligence comments on these changes, the category to which each comment belongs, the contribution rate of the software developer, the history of received feedback, and metrics regarding the impact on quality in accordance with the pull request data received through the code management platform (A), by establishing communication with this platform that enables the code changes made by software developers to be monitored and to be managed through pull requests.
3. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to run at least one artificial intelligence software thereon that automatically processes data and generates meaningful results.
4. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to establish communication with the database (2) and to record data in the database (2).
5. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to receive HTTP requests triggered by the processes of pull request creation and pull request updating performed by software developers on the code management platform (A).
6. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to automatically receive and process data related to pull request events via a developer notification connection tool, enabling these HTTP requests to be transmitted to the developer interface.
7. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to analyze code changes with metadata related to the source and target branch names of the pull request received with HTTP requests, the names of the modified files, which lines in each file have changed, and whether the file is new or an existing one.
8. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to establish the correct analysis context by determining the start and end boundaries of code blocks by applying syntax analysis to preserve the structural integrity of the code and identify the method to which each change belongs, by processing these specified line-based code differences.
9. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to evaluate, by using an artificial intelligence model running on the code management platform (A) to analyze the obtained methodbased code differences, the interpretation process of this model through a multilayered artificial intelligence architecture that manages this process in layers and provides quality control at every stage.
10. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to enable the technical quality of the code to be evaluated based on security, readability, and performance criteria within the scope of multi-layered artificial intelligence architecture, in order to step-by-step process and filter the comments obtained from the generative artificial intelligence model.
11. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to filter out ambiguous, unsupported, or unnecessary expressions that may appear in comments obtained from the generative artificial intelligence model.
12. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to check that the generated comments comply with current software standards and customer-specific rules.
13. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to enable only technically meaningful and truly necessary comments to remain in the system.
14. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to provide the developer with the overall context of the code changes made and to summarize all changes in an understandable manner.
15. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to generate comments that have been definitively validated and contextually appropriate by transferring the generated output to the next evaluation step, and to both transmit these comments directly to the relevant line of code through the code management platform (A) and to record them in the database (2).
16. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to record comments generated by the generative artificial intelligence model and stored in the system after their accuracy has been verified, along with the category information to which they belong, the identity of the developer the comment pertains to, the contribution rate, the frequency of receiving comments, historical feedback structures, and performance metrics reflecting their impact on code quality, into the database (2).
17. A system (1) according to any one of the preceding claims; characterized by the server (3) which is configured to analyze the development process of the developer by comparing current data with previously recorded comment content, contribution rates, method-based quality analyses, and time-series data in the database (2), and to provide team managers with meaningful outputs on development trends.
