Wizhut Research

Wizhut Research

Publications

Papers, and how to cite them

Every project in the lab publishes what it found on its own page. When the work is ready for readers outside the lab, it is written up as a paper: what the software is, the work it builds on, and how it is measured.

Papers

The work, written up.

Software paper · Journal of Open Source Software · In preparation

Kyori: string distance, similarity and ranking for autocomplete in JavaScript

Vassilios KarakoidasORCID 0009-0001-2263-3453 — Wizhut, Athens, Greece

The paper describes Kyori, the MIT-licensed JavaScript library behind the lab’s ranking study. It puts the classic edit distances — Levenshtein, Damerau–Levenshtein, Hamming — and the Jaro–Winkler similarity behind one contract of three functions, distance, similarity and rank, and adds a ranking cost of its own for ordering an autocomplete list once the typed text has filtered it. It traces each of the cost’s four rules to the prior work it borrows from, places the library among the metric collections and typeahead libraries already in use, and points to the evaluation that measures its order against eleven alternatives and two ablations.

String similarityEdit distanceAutocompleteQuery auto-completionRankingRecord linkageJavaScript
The evaluation The software on npm Source on GitHub

How to cite

Cite the software.

Until the paper is published, cite the software. The entry names the version every figure in the lab’s evaluation of Kyori was measured on; the paper’s own entry will join it when it has one.

The software — Kyori 0.5.0, the version the lab’s evaluation measured
@software{kyori_2026,
  author  = {Karakoidas, Vassilios},
  title   = {Kyori},
  version = {0.5.0},
  year    = {2026},
  license = {MIT},
  url     = {https://github.com/wizhut/kyori},
  note    = {JavaScript library, published on npm as @wizhut\_tech/kyori}
}

Wizhut Research

Every figure has a method behind it.

The projects these publications describe are free to use today: Kyori is one npm install away, MIT-licensed, and Promptivo scores a prompt in milliseconds with no model call.

The research projects Try Promptivo