Launching today
Judgie-AI
Multi-persona AI panel for code reviews & event evaluation
8 followers
Multi-persona AI panel for code reviews & event evaluation
8 followers
Judgie-AI is an open-source AI evaluation platform. Instead of a single score, it uses 5 expert AI personas (Entrepreneur, Engineer, UX Designer, PM, VC) to review projects from multiple angles. Features: - Multi-persona panel: Balanced & diverse reviews - Mid-event coaching: Tracks team growth with charts - "Objection!" feature: Re-deliberation loop - Admin AI chat: Q&A using source code - Bilingual & Custom Templates Deployable to Railway in minutes.
Hi Product Hunt community! 👋
I’m Yosuke Suzuki, Engineering Manager at Money Forward.
The Problem:
It all started when I was asked to judge an international internal hackathon. As a non-native English speaker, I dreaded the pressure of evaluating complex code, watching pitch videos, and writing meaningful, detailed feedback—all under a tight deadline and completely in English. I was terrified that my language barrier would compromise the quality of the judging.
The Solution:
I decided to hack the judging process itself and built Judgie-AI.
I realized that looking at a project through a single lens isn't enough. So, I built a panel of 5 distinct AI personas—a serial entrepreneur, principal engineer, UX designer, senior PM, and a VC investor. They don't just score; they generate multi-angle, bilingual feedback.
At the actual event, participant satisfaction averaged 4.62/5, and organizers rated it 4.67/5 for reducing judging burden. More importantly, it evolved from a mere "scoring machine" into a growth accelerator, helping teams iteratively improve their ideas during mid-event招consultations.
Going Global & Beyond Hackathons:
Realizing this was a universal challenge, I resolved the initial session management issues, rebuilt the UI, and released it as a robust OSS platform. Plus, it’s no longer just for hackathons—via JSON templates and GitHub Gist integration, you can easily load custom personas for startup pitches, hiring, or architecture reviews.
You can deploy it to Railway in just a few minutes. I would love to hear your thoughts, feedback, and what kind of evaluation templates you would build with it!
Let's change how we evaluate and accelerate project growth together! 🚀⚖️
I wrote a deeper post about why I built five AI judges—and why using multiple AIs makes human management more important, not less.
https://onehumanmanyais.substack.com/p/i-didnt-want-to-judge-a-hackathon
Where do you think human oversight matters most when AI evaluates people or products?
Curious how the five personas stay on track when they start disagreeing, is the deliberation loop fully visible to the team or just the final outcome?
@erafettinsabee
Thank you for great question.
At the moment, the five personas do not deliberate with one another. Each persona independently reviews the same submission through its own lens and produces its own score and feedback.
The team and human judges can see those different perspectives and where they disagree. There is no hidden deliberation loop or autonomous final verdict today.
Judgie-AI is not intended to replace human judges. It is designed to support judging and help participants improve their projects by surfacing multiple perspectives. The final interpretation and decision still belong to humans.
The five-persona debate setup is a clever way to surface blind spots a single reviewer would miss, and the objection loop caught a real flaw in my test project. Nice touch with the growth tracking charts for mid-hackathon coaching.
Curious how the 5 personas actually debate with each other under the hood, is it a fixed orchestration or can teams tune the discussion rounds and which persona leads the objection loop?