Jev Visual

# Jev Visual

English · [简体中文](README.zh-CN.md)

A small, runnable project for learning **vision-language model inference on Apple Silicon**. Use Qwen3.5-0.8B with MLX to answer multiple questions about one image: choose an option, judge yes/no, or score ordered levels. Includes a local browser UI, CLI and HTTP API.

> This project explores a Jev-like inference pattern for open multimodal language models. It avoids autoregressive structured generation by reusing shared multimodal context and directly scoring candidate outputs from model logits.
>
> This is an independent community implementation and does not claim to reproduce TypeSafe Jev's proprietary model architecture, RLCD training, calibration, or serving system.

## Visual game demos

<table>
  <tr>
    <th width="33%">AI sorting factory</th>
    <th width="33%">Breakout</th>
    <th width="33%">Camera gestures</th>
  </tr>
  <tr>
    <td valign="top">
      <video src="https://github.com/user-attachments/assets/c87c09d8-30d2-4392-b981-d0b5cf1879ae" controls width="100%"></video>
      <p>Classify conveyor objects and choose a sorting lane.</p>
      <a href="https://github.com/user-attachments/assets/c87c09d8-30d2-4392-b981-d0b5cf1879ae">Watch video</a>
    </td>
    <td valign="top">
      <video src="https://github.com/user-attachments/assets/a642a6d6-c138-4c02-8378-f2981be22aad" controls width="100%"></video>
      <p>Locate the ball in five regions to control the paddle.</p>
      <a href="https://github.com/user-attachments/assets/a642a6d6-c138-4c02-8378-f2981be22aad">Watch video</a>
    </td>
    <td valign="top">
      <video src="https://github.com/user-attachments/assets/b0042194-54f0-4fc4-9534-e31ec001aad8" controls width="100%"></video>
      <p>Recognize hand gestures from camera frames to control pa