Jev in production › SQL and data pipelines
jlink
Vendor claims Jev-based record linkage matched 171,354 firm-name pairs for economists at $2.95 total, about 250 pairs per second, from Python, Stata and R (author).
- Use
- SQL and data pipelines
- Industry
- Education and research
- Form
- Open-source tool
- Stage
- Announced
- Listed
- 2026-09-25
- Found via
- github
- Repository
- keltokhy/jlink
- Stars
- 6
- Forks
- 1
- Last push
- 2026-10-02
- Language
- Python
- License
- MIT
The README opens with
Record linkage where you write the match rule in plain English.
String distance links "Acme Widgets Inc." to "ACME WIDGETS, INC". It does not link "IBM" to "International Business Machines", and it cannot know that you want a subsidiary kept apart from its parent. A research assistant can do both, slowly. jlink asks Jev, a decision model from TypeSafe, the question a research assistant would answer: given these two records and this rule, are they the same firm? Jev returns a probability in about a fifth of a second for about a thousandth of a cent. In the benchmarks below, 171,354 pairs cost $2.95 in total, at about 250 pairs a second. With no labels, jlink beats LinkTransformer's zero-shot models on eleven of twelve standard benchmarks and its...
Badge
For the project's own README, linking back here:
Also used for sql and data pipelines
- JevSDSQL $0.389
- grev sub-cent cost per few thousand lines
- Jev IDS 7.7x
- jev-harness (AntonioCoppe) 1.3 seconds
- jev-ultralightspeed 41% less money
- signal-weave 130/144 exact outcome decisions
- jevgrep mean F1 of 0.90
- Vibefilter 1,000 rows in about a second