We built a tiny Jev on one PC. Only then did we see what it is really for

What Jev is

Jev is a decision-only AI that raised $40M. It does not write. You give it a situation and a fixed set of options, and it returns one choice, a probability for each option, and a confidence score.

We borrowed the same input and output shape and narrowed it to one job. The video above walks through it, from the big picture to the details. If you only want the big picture, the first few minutes are enough.

Three things we learned

1. Narrow the scope, and it gets small and fast. A typical decision took about 2.4 ms on a single PC. This is not a contest with Jev's reported 70 to 500 ms: theirs includes a round trip to a cloud server, ours is the decision alone.
2. Once you fix the options and ask for a confidence score, anyone can start designing one.
3. Speed is not enough. Test on known answers first, and send low-confidence cases to a person.

Where it fits

In front of the rule gates and human approval that agents already need, taking the repeated decisions off their plate. The gate and the person stay.

The paper is on its way. The full process will follow.

What's next

If enough people are interested, we'll open-source our small judge on GitHub. If not, we'll keep improving it and share it when there is something worth sharing. If you're interested, drop us a line through contact.

The Korean original

Watch elsewhere

• Full video in Korean, with English subtitles (YouTube)
• 60-second short (YouTube)
• The LinkedIn post
• The Korean version on this site

The voice in this video is an English voice-over; the speaker is Korean.