Nyehash AI

Kirundi, kept legible to machines.

Nyehash AI builds datasets, models, and morphology-aware tools for Kirundi — an agglutinative, tonal Bantu language spoken across Burundi and beyond.

Ishirahamwe ryo gukingira Ikirundi mu mashini nyabwonko

Why this exists

Most NLP tooling is built for isolating, non-agglutinative languages. Kirundi packs a noun's class, a verb's subject, tense, object, and voice into single inflected words — a plain subword tokenizer breaks that structure apart at meaningless boundaries instead of meaningful ones.

We build morphology-aware tools, open datasets, and models that respect that structure, so Kirundi text stays usable — for translation, search, education, and archival work — instead of being flattened into noise.

A quick example

What we build

• Corpora — collecting and cleaning Kirundi text (oral literature, news, dictionaries) into open, citable datasets.
• Morphology tools — rule-based analyzers and tokenizers built on Kirundi's noun-class and concord system, not borrowed defaults.
• Language models — fine-tunes and small models trained with Kirundi's grammar in mind, for translation and generation.

📦 Datasets · 🤖 Models · ✉️ nyehashi@gmail.com

© Nyehash AI — digital preservation of the Kirundi language