Overview & result
chenmingtang830 shows jevgraph, built on Jev: Knowledge graphs built from typed relation decisions, with the evidence kept.
From the original post
jevgraph by @chenmingtang830: Knowledge graphs built from typed relation decisions, with the evidence kept.
How Jev fits in the loop
- Documents are processed against a fixed or slowly changing schema, in bounded stages instead of one open-ended "prompt to triples" step.
- For each candidate pair of entities, a typed decision selects one relation from the schema's explicit list.
- Each edge keeps its source evidence.
- The output is a candidate knowledge graph that can be inspected stage by stage.
How to reproduce
- Review the source documentation and repository at https://github.com/chenmingtang830/jevgraph.
- Configure environment with TypeSafe Jev / System One API keys.
- Run test prompts against the decision endpoints.
Why this build matters
Shows Jev in practical use for research and data analysis.
Limitations
- The README says it does not claim model output is true, approved or safe for consequential reuse.
- The primary benchmark measures one operation only: relation identification on 160 pinned FewRel cases.
- Jev's cost in the benchmark is an illustrative list-price equivalent, not a provider receipt.