jevbooks

← All projects

reflex

kshetrajna12/reflex

A tiny decision model: state plus typed questions, calibrated probabilities.

What it is

An open re-creation of Jev, TypeSafe's System One model, built on Qwen3.5-4B. It answers typed questions about a state (text or image) with calibrated probabilities, for developers who need fast structured judgments instead of prose.

How it uses Jev

Jev is the model reflex re-creates. Given a state and typed questions (choice, score, noul), reflex returns calibrated probabilities in one forward pass, about 200 ms. Your code sets thresholds on those numbers, e.g. auto-route if above 90%.

Primitives:choicescorenoul

Technique worth stealing

Readout: lettered yes/no answer, Evidence / Criterion framing, averaging two distinct option orders.

Try it

git checkout stable && uv run reflex-serve --stable on a 16 GB CUDA GPU; or try the WebGPU demo.

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.84
  • Shows a System One patternyes0.78
  • Handles uncertaintyno0.06
  • Measuredno0.09
  • Runnableyes0.84
  • Worth recommendingno0.38
  • Model replicayes0.96
  • Problem scopescore on a 0–2 scale0.91
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 21 Sept 2026.