🔗 GitHub: https://github.com/krishagarwal314/CodeJury
🌐 Website / Demo: https://krishagarwal314.github.io/CodeJury/
I strongly recommend visiting the website first. It has the visual explanation of how CodeJury works and will give you a much better understanding of the project than this blog post.
The GitHub repository has the implementation, architecture, prompts, and everything else if you want to dig deeper.
I've been using AI coding tools more and more, and one thing kept bothering me:
AI is getting extremely good at writing code.
But it is also extremely good at writing code that looks correct.
That's what led me to build CodeJury.
The idea is simple: instead of trusting a single AI model to review AI-generated code, CodeJury gets two independent reviewers from different model families to inspect it.
Your Code
│
┌────────┴────────┐
▼ ▼
Reviewer A Reviewer B
│ │
└────────┬────────┘
▼
Independent
Reviews
The goal isn't to replace human code review or claim that multiple LLMs magically produce perfect reviews.
It's to introduce independent opinions into the process.
If one model misses something, there's a chance the other catches it.
If both reviewers agree, that's useful signal.
And if they disagree, that's arguably even more interesting — it tells you where the code deserves another look.
Why different models?
Running the same model twice doesn't necessarily give you much independence.
Different model families can have different strengths, weaknesses, and failure modes.
So CodeJury deliberately uses model diversity as part of the review process.
And what about cost?
I didn't want this to become:
"Run three giant LLMs with your entire codebase and pay a fortune for every review."
The system keeps the review context focused and avoids unnecessary context being sent to the models.
The idea is to make getting a second opinion cheap enough that it can actually become part of an AI-assisted development workflow.
This is still an experiment, and there's a lot I'd like to explore further — better reviewer specialization, measuring agreement between models, evaluating review quality systematically, and understanding when model diversity actually improves results.
If you're interested in AI coding agents, LLM evaluation, or automated code review, I'd love for you to check it out.
Start here
🌐 Website:
https://krishagarwal314.github.io/CodeJury/
💻 GitHub:
https://github.com/krishagarwal314/CodeJury/
The website is the best place to start — the GitHub repo is there if you want to go deeper.
Top comments (0)