jevtree
Builds auditable decision trees and runnable decision SOPs from data and a natural-language goal via LLM features and information gain.
Overview
jevtree, documented mainly in Chinese with an English README, connects structured tables, text or folders with a natural-language goal and produces an auditable decision tree and a runnable DecisionSOP. An LLM first normalises the data into a feature table; a decision tree is then grown by information gain, and the tool exports tree.json, sop.json and a bilingual narrative. The SOP can be run on new observations with an optional trace showing the question-and-answer path taken.
It also includes an active feature acquisition evaluation under hard budgets; on the MiniBooNE dataset the README reports accuracy of 0.824 at a budget of five features for its discriminative information-gain policy against 0.746 for random selection. The list describes it as an LLM-authored decision SOP generator for the Jev System One model; the README excerpt does not show Jev calls directly, and its LLM is configured through an OpenAI-style endpoint. It suits analysts wanting transparent decision procedures.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which branch applies to this observation at each SOP node (e.g. approve the loan or not)?(inferred) | — |
How it works
- Provide data and a natural-language goal
- LLM normalises inputs into a feature table
- Grow a decision tree by information gain
- Export tree, runnable SOP and narrative
- Run the SOP on new observations with an optional trace
Strengths
- Auditable decision paths
- No gradient training required
- Budgeted feature acquisition evaluation
Caveats
- Jev's role is not shown in the README excerpt
- Documentation primarily in Chinese
- Requires an LLM API key
Getting started
pip install -e ., set JEVTREE_LLM_API_KEY in .env, then jevtree decide --data examples/data/bring_your_csv/loan_approve.csv --goal "..." --out results/loan_demoAt a glance
- Type
- Project
- Status
- research
- Language
- Python
- Runtime
- CLI (Python 3.11+)
- Has code
- yes
Use cases and integrations
How curated lists describe it
LLM-authored decision SOP generator for Jev System One model.
hellogumbo/awesome-jev
Frequently asked questions
What does jevtree produce?
A decision tree, a runnable decision SOP and a bilingual narrative from your data and goal.
Is it trained with gradients?
No; the tree grows by information gain after LLM normalisation.
Where does Jev come in?
The listing calls it an SOP generator for Jev; the README excerpt focuses on the LLM and tree pipeline.
Related projects
Standalone Node studio for building and testing workflows of typed Jev judgments with code-owned decisions, plus games and simulations.
ProjectbetaJavaScriptSelf-hosted Yao Agents platform for agents, workspaces and task boards across devices, which announces Jev support.
ProjectbetaCommand-line tool and SDK that answers typed questions about text with true/false, a label or a number, for scripts and evals.
ProjectbetaRustReusable project context and specialist workflows for existing AI tools, with an addon that offloads bounded decisions to Jev.
ProjectproductionTypeScriptVerifiable coding-agent runtime built on checkable execution graphs, listed with an optional Jev provider for bounded choices like routing.
ServiceproductionTypeScriptAutoGPT agent platform, listed as adding Jev decision blocks for typed routing, filtering, scoring and confidence-gated next actions.
ProjectproductionPython