Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-dev

Japanese demo judging the same utterance with Jev and an LLM to compare emotion-shift outputs and speed side by side.

Serviceprototype
View on GitHub ↗

Overview

jev × LLM judgement viewer is a Japanese-language demo that sends the same user utterance to Jev and to an LLM (via Vercel AI Gateway) and shows the difference in their judgements and their speed on one screen. Each judgement updates a separate emotional state managed by affectus, an eight-axis emotion model, and the two states diverge over a conversation, shown on Plutchik wheels.

Both sides then generate a reply with the same model (Haiku 4.5), persona and instructions, differing only in the emotional state they carry, so judgement differences show up as differences in tone. The screen shows per-axis deltas, a total divergence score, judgement and generation times, and the raw judgement JSON. It requires macOS, Node 24+ and a Vercel AI Gateway key, and the distribution bundles the affectus binary and dependencies.

What the decision model decides

TypeQuestionAnswers
ScoreHow much does this utterance move each of the eight emotion axes?(inferred)—

How it works

  1. User sends an utterance
  2. Jev and the LLM each judge emotion deltas from their own current state
  3. Update two separate affectus states and draw Plutchik wheels
  4. Generate a reply from each state with the same model and compare

Strengths

  • Side-by-side view of judgement differences and latency
  • Shows how judgement differences change downstream replies
  • Bundled binary; no Go or network installs needed

Caveats

  • macOS only
  • Fair comparison only on the first turn, since states diverge afterwards
  • Japanese UI and docs

Getting started

`bash scripts/setup.sh`, add the API key to .env.local, then `node --env-file=.env.local src/server.ts` and open http://localhost:8787.

At a glance

Type
Service
Status
prototype
Language
TypeScript
Runtime
Node 24 local server
Has code
yes

Use cases and integrations

How curated lists describe it

  • 同じ発言を jev と LLM の両方に判定させ、感情の変動値のズレと応答速度を1画面で見比べるデモ（affectus + Vercel AI Gateway）.

    hellogumbo/awesome-jev

Frequently asked questions

What does Jev judge?

How a user utterance shifts each of eight emotion axes.

How is the comparison kept fair?

Both sides start from the same state; after the first turn their states diverge by design.

What do I need?

macOS, Node 24+ and a Vercel AI Gateway API key.

Related projects

  • jev-arena★ 122

    Side-by-side tool comparing Jev and DeepSeek on labelling the same comments for relevance, sentiment and intent, with replay and reports.

    ProjectbetaTypeScript
  • jev-dimabsa

    Jev applied to SemEval-2026 DimABSA dimensional sentiment tasks using only typed primitives, no generation or fine-tuning.

    ProjectresearchPython
  • jevtest (joshhu)

    Chinese 'emotion lie detector' demo comparing Jev's Choice, Score and Noul judgments on a message with a general LLM.

    CLIprototypePython
  • candidate-experience-benchmark

    Benchmark comparing Jev, other decision models and LLMs on four judgments over 60 synthetic candidate-experience reviews.

    Benchmarkresearch
  • Jev vs. ML

    Benchmark of Jev 1.13.0 against 11 classical classification pipelines on eight datasets under a published protocol.

    BenchmarkresearchPython
  • jevbench (dhruvmehra)

    Small reproducible benchmark of Jev against LLMs, fine-tuned BERT, Laya and zero-shot NLI on SST-2, AG News and Banking77.

    BenchmarkresearchPython