Kashmir AI Research
Seven million speakers. Almost no machine that understands them.
Kashmir AI Research builds the missing pieces for Kashmiri: a corpus, a benchmark, and a place for native speakers to judge the results. It is led by Faizan Ayoub, who also designed and built AmanahVale.
- people speak Kashmiri
- 7M+people speak Kashmiri
- Kashmiri–English sentence pairs, aligned and filtered
- 124,102Kashmiri–English sentence pairs, aligned and filtered
- resources catalogued, 2003–2026
- 115resources catalogued, 2003–2026
Best Paper, ICRTC-2026 — Anatomy of Decoder-Only LLM Failure in Low-Resource Machine Translation
What it builds
A corpus
Sentences in Kashmiri and English, lined up pair by pair — the raw material any translation model has to learn from.
The datasetA benchmark
Translation models tested on the same sentences, so anyone can see which ones work for Kashmiri and which do not.
The researchNative speakers as judges
A score from a machine is not enough. People who speak Kashmiri rate the translations themselves.
Try the translatorFrom the classroom
Why it belongs with the lessons
In Class 10, students learn how a machine reads language — and find that it barely reads theirs. Kashmiri has far less written text than English for a model to learn from. This research is working on exactly that gap, here, in the valley.
So students do not only learn about AI in Kashmiri — they can help build it. Speaking the language is something no model can do for them.See the lessons in the library
How a class takes part
- 1In class, students write short Kashmiri sentences with their English meanings, and check translations a model has made — on paper or in a class sheet.
- 2Their teacher reviews the work and sends it in for the class. Students never make an account on the research site, and nothing they send names them.
- 3A parent agrees first, because the sentences may become part of an open dataset. The school and the class are credited — never a child by name.
Teachers, parents and other adults who speak Kashmiri can rate translations on the research site themselves.
Learn AI. Keep the trust.