@Nikerabbit hey! Please check again. It should be solved.

I'm in Wikimania. This morning I should have some time. If I find a plug to connect my pc I'll have a look at this.

Hi @Nikerabbit @abi_
Really, thanks for your work and commitment. During this period, I have a backlog of work, but I'll get back to it ASAP.

The idea is to create a user script that modifies the standard Meta-Wiki (www.metawiki.org) or other project interface, adding a button like "add translation tag" that calls the gadget and modifies the page content to add translation tags.

LGTM, thanks!

You can find a long list of supported projects here:
https://meta.wikimedia.org/wiki/Help:Interwiki_linking_on_Wikimedia_wikis
https://www.mediawiki.org/wiki/Manual:Interwiki

One way to implement this is with FastAPI decorators (from fastapi import FastAPI), which automatically generate a description of available endpoints in the Swagger UI.
You can find an example here: https://github.com/ftosoni/mediawiki-code2code-search/blob/ff84c9e399862cb976478b6859c1fc995d50feb9/app.py#L335

See, e.g.,

• https://accessibility-checker.toolforge.org/docs
• https://code2codesearch.toolforge.org/docs

An idea could be to inherit the interface/frontend from https://accessibility-checker.toolforge.org/ which is also built according to the Codex design system

Also, it should support parsing of wikilinks to other wikis, such as French Wikivoyage, English Wikiquote, Hindi Wikisource, and Wikidata.

Also refer to https://www.wikidata.org/wiki/Wikidata:List_of_Wikimedia_tools_with_Wikidata_item
Licences, etc., are all structured data that can be easily integrated into the tools' Wikidata items.

Hi! I uploaded the slides on Commons: https://commons.wikimedia.org/wiki/File:Mediawiki-code2codesearch-wmhackathon2026-slides.pdf

Heya! I don't think it's accurate to say that the tool «sends the input to at least a third-party AI service».
WMDE staff could correct me if I'm wrong. My understanding is that graph embeddings are computed locally on WMDE/WMF servers using Jina AI models (likely downloaded from Hugging Face). Then the resulting embeddings/vectors are used to build a Vector DB, which, in this case, is an IBM-proprietary solution called DataStax, built on Apache Cassandra. This is done offline. Clearly, the query itself (i.e., the text the user types into the search box) needs to be vectorised as well. This is done online using the same Jina AI model. After the query is vectorised, the resulting embedding is sent out to IBM servers where the vector DB is hosted to perform an (approximate) nearest neighbour (ANN) search and retrieve closest embeddings, corresponding to "similar" Wikidata Q-items.

The fork has been reviewed and merged on GitHub. See https://github.com/indictechcom/translatable-wikitext-converter

Could you provide the exact input that generated the wrong output? @Ata

Hi! @Ata and the others :) Sorry for the delay.
I just gave a quick look and I couldn't reproduce the error.
If I go to https://translatetagger.toolforge.org/convert
and insert the Input Wikitext:

This is a testing example.
*What: Enter the title of the event
*Register your interest, using your signature (4 tildes):

The Translatable Wikitext Output is:

<translate>This is a testing example.</translate>
* <translate>What: Enter the title of the event</translate>
* <translate>Register your interest, using your signature (4 tildes):</translate>

In this case, the parsing is correct.

I tried to address this in a fork: https://github.com/super-nabla/translatable-wikitext-converter.
I also posted this in T393170

Well done! Nice tool. @Gopavasanth, I forked your project and tried to add support for "tvar"s:
https://github.com/super-nabla/translatable-wikitext-converter.
I'm quite a newbie here, so please let me know what you think and if there's anything I should fix.

Thank you, Izno!