An AI code review tool with multiple personalities—from brutal roasts to supportive mentor feedback—so developers can inspect their code from different angles.
Timeline
Jul 22, 2025 – Aug 19, 2025
Role
Fullstack Development
Focus
AI-powered code review
Primary Stack
React.js
TypeScript
Tailwind CSS
OpenAI API
Zustand
VitestCase overview
Built an AI-powered code review experience where developers paste code, pick a review persona, and get structured feedback in seconds.
Why I Built It
Most code review tools feel either too dry or too generic. I wanted something that still teaches best practices, but in a voice that matches the mood—whether that’s brutal roasting or a calm mentor tone.
CodeRoaster experiments with how tone affects how feedback is received. The same code can be reviewed as a roast, as a supportive mentor, or as a strict professional reviewer.
Review Engine & UX
The core flow is simple: paste code, choose a persona, and hit review. Behind that, prompts are tailored per persona so the output stays consistent in style while still being technically useful.
The interface is built around readability: monospace code area, clearly separated feedback sections, and persona switches that make it easy to rerun the same snippet with different review styles.
What I Learned
Tuning prompts for consistent tone taught me how sensitive AI behavior is to small wording changes, especially when you want personality without losing technical depth.
Designing the flow also reinforced how important it is to keep friction low for developers: the less setup needed, the more likely they are to actually run a review before pushing code.
