---
title: Three tools I built with an AI that only judges
description: "Jev doesn't write a word. It picks an answer and tells you how sure it is. Three tools for designers, each one running in seconds for less than a cent."
date: 2026-10-12
tags: [ai, design, build-in-public, jev, video]
url: "https://mahdif.com/blog/three-tools-i-built-with-an-ai-that-only-judges/"
---

# Three tools I built with an AI that only judges

Jev doesn't write a word. It picks an answer and tells you how sure it is. Three tools for designers, each one running in seconds for less than a cent.

<figure style="margin:0 0 2.75rem;" data-wide>
<div style="position:relative;width:100%;aspect-ratio:16/9;border:1px solid var(--border);border-radius:var(--radius-lg);overflow:hidden;background:var(--card);"><iframe src="https://www.youtube-nocookie.com/embed/U3FMv4_TqrA" title="Three tools I built with an AI that only judges" style="position:absolute;inset:0;width:100%;height:100%;border:0;" allow="accelerometer; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen loading="lazy"></iframe></div>
<figcaption>7 minutes. The copy grader starts at 2:50.</figcaption>
</figure>

"Can we make this pop more?" Every designer has read that comment and had no idea what to do with it.

So I built a tool that checks feedback before it reaches you. Vague comments go back to the person who wrote them. All 8 on a test page took 0.6 seconds.

It runs on Jev, a new model from TypeSafe. It doesn't write anything. You give it something to look at and the only answers it may give, and it picks one and says how sure it is.

I built two more with it. One graded all 318 strings in Peeko in 18 seconds, and 15 of them came back as lost for a first-time user. The other searches lvly.design by feel and ranks all 127 sites in 7 seconds. That one is still a prototype.

Each run cost less than a cent. All of this was possible before. It was just too slow and too expensive to run on everything.
