تعزيز جودة الشيفرة على نطاق واسع: uReview من أوبر، حل مراجعة الشيفرة باستخدام الذكاء الاصطناعي

In the fast-paced world of software development, code reviews play a crucial role in maintaining the quality and reliability of codebases. However, with the rise of complex services and AI-assisted development, traditional peer reviews face significant challenges. Uber has introduced uReview, an AI-driven code review platform designed to address these challenges and enhance the efficiency of the code review process.

Introduction

Code reviews are essential for ensuring the reliability and safety of software. At Uber, where tens of thousands of code changes occur weekly, traditional review methods are increasingly strained. uReview aims to alleviate this burden by providing a scalable solution that integrates AI into the code review process, thereby improving the quality of feedback and reducing the workload on human reviewers.

The Challenges of Traditional Code Reviews

• Overloaded Reviewers: The volume of code changes makes it difficult for reviewers to keep up.
• Limited Time: Reviewers often lack the time to identify subtle bugs and enforce best practices.
• Missed Errors: Traditional methods can lead to production incidents and slow release cycles.

Introducing uReview

uReview serves as a second AI reviewer, enhancing the code review process through its core component, Commenter. This modular, multi-stage GenAI system focuses on:

• Identifying Bugs: Detects functional bugs, error handling issues, and security vulnerabilities.
• Proposing Fixes: Suggests code changes based on comments from both humans and AI.

How uReview Works

The architecture of uReview is designed to streamline the code review process through several stages:

1. Ingestion and Preprocessing: Filters out low-signal files and builds structured prompts from relevant code context.
2. Comment Generation: Utilizes a pluggable assistant framework, with specialized assistants for different types of issues:
   • Standard Assistant: Detects bugs and logic flaws.
   • Best Practices Assistant: Enforces coding conventions.
   • AppSec Assistant: Focuses on security vulnerabilities.
3. Post-Processing and Quality Filtering: Implements multi-layered filtering to ensure high-quality, actionable feedback.
4. Comment Delivery: Posts validated comments directly to the code review platform, allowing developers to rate their usefulness.

Performance and Impact

• High Usefulness Rate: uReview maintains a usefulness rate above 75%, with 65% of comments being addressed by developers.
• Time Savings: Processes over 10,000 commits weekly, saving approximately 1,500 hours of developer time every week.
• Empirical Evaluation: Benchmarks against third-party models show uReview’s superior performance in identifying bugs and providing valuable insights.

Lessons Learned

The development of uReview has provided key insights into building scalable AI tools:

• Prioritize Precision: Quality of comments is more important than quantity.
• Incorporate Feedback: Real-time feedback from developers is crucial for system tuning.
• Implement Guardrails: Multi-stage prompting and filtering are essential to reduce false positives.

Future Directions

Looking ahead, Uber plans to expand uReview’s capabilities:

• Support for More Review Categories: Including performance and test coverage.
• Enhanced Contextual Understanding: Developing tools to better assess code risks and dependencies.

Conclusion

uReview represents a significant advancement in Uber’s approach to code quality. By leveraging AI to complement human insight, uReview not only enhances engineering productivity but also allows developers to focus on higher-order tasks. With its proven effectiveness and ongoing improvements, uReview sets a new standard for AI-assisted code reviews in the software development industry.

Source: Uber Blog