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

typesafe-jev-model-use-cases

Small Python demos of Jev: parallel Noul ticket checks with measured speedup, and a three-way Choice citation and claim checker.

Appprototype
View on GitHub ↗

Overview

This repository has small runnable demos of Jev. The first judges 16 sample support tickets ("is this about billing?") sequentially and then concurrently with a thread pool, printing wall-clock time and token use. The author reports a consistent 4.5 to 5 times speedup from parallel calls at no extra token cost, with matching results, and advises parallelizing batches rather than looping.

The second is a citation checker: 8 chatbot-style claims are checked against 3 short policy documents with a three-way Choice of supports, contradicts or says_nothing, rather than yes/no, so wrong claims are distinguished from unsupported ones. Low-confidence verdicts are flagged for human review. Both follow patterns from TypeSafe's cookbooks.

What the decision model decides

TypeQuestionAnswers
Yes / noIs this ticket about billing?yes · no
ChoiceDoes the source support, contradict, or say nothing about this claim?supports · contradicts · says_nothing

How it works

  1. Install requirements and set TYPESAFE_API_KEY.
  2. Run run.py to compare sequential and parallel Noul calls.
  3. Run citation_check.py to verify claims against sources.
  4. Review flagged low-confidence or unsupported claims.

Strengths

  • Measured parallel speedup of about 4.5 to 5 times.
  • Three-way choice separates wrong from unsupported claims.

Caveats

  • Small sample datasets.
  • Demos, not a library.

Getting started

pip install -r requirements.txt, set TYPESAFE_API_KEY, then python run.py or python citation_check.py

At a glance

Type
App
Status
prototype
Language
Python
Runtime
scripts
Has code
yes
Stars
1

Use cases and integrations

How curated lists describe it

  • Runnable demos of TypeSafe's System One model (Jev) — parallel Noul judgments and a Choice-based citation/claim checker

    wh000wh000/awesome-jev-live

Frequently asked questions

Why a three-way choice for citations?

To separate claims that contradict the source from claims the source simply does not support.

How much faster is parallel calling?

About 4.5 to 5 times in the author's test.

Is the model deterministic?

No; the README notes occasional mismatches between runs are expected.

Related projects

  • keeping-an-npc-on-quest

    Educational code where a local LLM voices a game NPC and Jev classifies player messages and checks every line for spoilers.

    ProjectprototypePython
  • Support-Ticket-Ai-Agent-with-RAG-Guardrails-and-Jev

    LangGraph student-support agent with PII masking and hybrid RAG, where Jev verifies drafts before auto-reply or human review.

    ProjectprototypePython
  • jev-qa-demos

    TypeScript demos for testers: Jev triages CI test failures, gates test-agent commands and RAG answers, across gateway, Ollama and Foundry.

    LibraryprototypeTypeScript
  • Double-checking citations

    Official TypeSafe cookbook example for checking whether source context actually supports a claim.

    Articleproduction
  • jev-starter-examples

    Two beginner Python apps using Jev: a message sorter and a support ticket router that send uncertain items to human review.

    ProjectprototypePython
  • Jevidence

    Python sandbox turning Jev issue-category and reproduction judgments into testable routing proposals, with offline replay.

    ProjectprototypePython