What it is
A personal experimentation demo that runs a simulated fraud interceptor over synthetic transactions, comparing a fast typed-decision model (Jev) against an LLM (Gemini) in parallel on latency, token usage and decisions. Not a benchmark.
How it uses Jev
Jev receives each synthetic transaction and returns a typed decision: decision (APROBAR_DIRECTO, SOLICITAR_2FA, BLOQUEO_PREVENTIVO) with confidence and per-option probabilities, plus is_fraud as a noul probability between 0 and 1. The demo compares this against Gemini's decision, report and token usage.
Primitives:choicenoul
Technique worth stealing
Run a fast typed-decision model in parallel with a deliberative LLM and compare decisions, latency and tokens.
Try it
git clone repo, npm install, cp .env.example .env, add keys, npm run dev; open http://localhost:3000.
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-centricyes0.62
- Shows a System One patternunclear0.52
- Handles uncertaintyno0.18
- Measuredno0.08
- Runnableno0.18
- Worth recommendingno0.27
- Model replicano0.07
- Problem scopescore on a 0–2 scale0.42
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.