Jev directory
CommunityJev Demos & Experiments
Jev Playground — taiki00wo
A Japanese-language experiment workspace that asks Jev, through typesafe-sdk-python's jev-latest, why a dummy restaurant's sales fell, and compares Jev against a local Ollama model on accuracy, speed, answer variance, confidence, and cost.
- Stars
- 0
- Reading time
- 2 min
- Published by
- Community
- Added
- Nov 2026
This is one of several Jev listings for the same project. The canonical entry is Jev Playground (MoonBit).
Highlights
- Asks Jev via typesafe-sdk-python's jev-latest about causes of a restaurant sales decline.
- Store data is labeled dummy data; no real store records are used.
- experiments.py walks four stages, from clear-answer data to passing computed change rates.
- compare.py measures accuracy, speed, answer variance, confidence, and cost against Ollama.
- Results are written as JSON under results/, which is excluded from Git.
Questions about this entry
Short answers drawn from the linked project or page.
Run uv sync, copy .env.example to .env, and put your TypeSafe AI API key in TYPESAFE_API_KEY. For the LLM comparison, start Ollama and pull the model, by default llama3:8b, with ollama pull llama3:8b.
Only for the Jev side. Passing --only llm runs just the Ollama comparison and needs no API key; running both sides, or --only jev, requires the key set in .env.
It uses dummy store data that carries correct labels. The repository states the data is not from any real store, so results are illustrative rather than drawn from live business records.
It compares Jev and an Ollama model on the same cause-judgment task, reporting accuracy, speed, how much answers vary across repeats, confidence, and cost. Results are saved as JSON in results/.
The script changes how Jev is asked in four steps: clear-answer data using Choice, Score, and Noul; mixed causes with and without a comparison baseline; per-cause Noul questions with observations mixed in; and passing a computed change rate while asking only for the judgment.
Yes. compare.py accepts --llm-model for the Ollama model, --per-cause for stores per cause (default 5, total four times that), --repeats for repeated judgments (default 3), --limit to cap records, and --seed for the dummy-data random seed (default 42).
The repository is MIT licensed. It is a personal experiment workspace rather than an official TypeSafe AI project, and it depends on the typesafe-sdk-python client for its Jev calls.
More Jev resources like this
A browser playground for S1-CED, a small decision model you give a piece of text plus typed questions — yes/no, which-of, or a 0-3 score — and it answers with a probability per option and whether it is confident enough to act or a human should check. Runs on CPU; no GPU or account needed.
Jev Demos & Experiments#community#github#python
Community
ai-yatra/s1ced-playgroundPython
ai-yatra/s1ced-playgroundPythonA Python demo that screens PubMed abstracts for one systematic review question using Jev. Six narrow questions per abstract go to Jev in a single call, with rule-based tiers and confidence gates deciding include, exclude, or human review.
Jev Demos & Experiments#demo#python#madewithjev
Communityanondo1969/jev-evidence-triage
A small proof-of-concept app that runs local Laya and external System One-family APIs in one judgment UI: it executes typed questions against a state and compares choice probabilities, yes/no probabilities and ordinal scores across providers.
Jev Demos & Experiments#community#github#python
Community
keispace/systemonePython
keispace/systemonePythonA MuJoCo city simulation where one, two or four camera-observed drones pursue a target: each role asks Jev role-specific typed questions while deterministic mission navigation, safety guards and motor control run independently at 50 Hz.
Jev Demos & Experiments#community#github#python
Community
YongLD/Multi-UAV-JEVPython
YongLD/Multi-UAV-JEVPythonAn open-source browser 3D character split between two models: GPT-Live-1 talks with you over WebRTC while Jev reads the live transcript and picks Annie's face, gestures and body language in about 120 ms, with code gates on every answer.
Jev Demos & Experiments#community#github#python
Community
rehan-remade/open-anniePython
rehan-remade/open-anniePythonA Chinese-language e-commerce customer-service demo on FastAPI, LangGraph and React: adaptive retrieval routing with hybrid BM25-plus-vector search, and intent classification switchable between TypeSafe Jev's structured choice interface and DeepSeek.
Jev Demos & Experiments#community#github#python
Community
gbh1234/Adaptive-ragPython
gbh1234/Adaptive-ragPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.