CodeRev

scoring system
detailed review along with severity tag and file path
beginner-mode demo
clerk authentication

Inspiration

Code reviews are one of the most valuable parts of software development, but they’re also time-consuming and often unavailable when you need quick feedback. While building projects, I noticed that most existing tools either focus on static analysis or provide overwhelming feedback without context. I wanted to explore whether an AI agent could perform practical, human-like code reviews that focus on what actually matters: bugs, security issues, and maintainability.

This project was inspired by the idea of making high-quality code review more accessible to developers of all experience levels, especially when fast, actionable feedback is needed.

What it does

CodeRev is a general-purpose AI tool that reviews real project code and provides structured, actionable feedback.

It analyzes code changes using Git context and filesystem awareness, then highlights:

Potential bugs and logical issues
Security vulnerabilities and unsafe patterns
Code quality and readability concerns
Maintainability issues across files

The tool also generates a transparent code health score based on four factors — bug risk, security, code quality, and maintainability — helping developers quickly identify areas that need attention. For beginners, it can explain critical issues in simple, easy-to-understand language.

How we built it

The project is built around an AI agent architecture powered by the Gemini API. Instead of analyzing isolated snippets, the agent uses MCP servers to understand real project context.

Gemini API for reasoning over code and generating structured review feedback
MCP Git server to review diffs and recent changes, similar to a pull request
A lightweight web interface for submitting code and viewing results
A scoring layer that converts structured findings into explainable metrics

This approach allows the reviewer to behave more like a human code reviewer rather than a simple text analyzer.

Challenges we ran into

One of the main challenges was designing prompts that produce useful, structured feedback without overwhelming the user. Striking the right balance between depth and clarity required multiple iterations.

Another challenge was ensuring that the scoring system remained transparent and defensible. Rather than relying on a single AI-generated number, the score had to be derived from clear, observable findings so users could trust and understand it.

Accomplishments that we're proud of

Building a working AI agent that performs context-aware code reviews
Designing a transparent scoring system instead of an arbitrary AI score
Making code review feedback more accessible through beginner-friendly explanations

What we learned

This project reinforced the importance of prompt design and structure when building AI-powered tools. Small changes in how instructions are framed can significantly impact the usefulness of AI output.

I also learned how agent-based architectures, combined with tools like Git and filesystem access, can dramatically improve the quality of AI reasoning compared to isolated inputs.