KhasiBERT is the first open-source AI language model trained exclusively on Khasi-language corpora. Developed by MWire Labs, it supports civic NLP tasks such as translation, summarization, and search, and is designed for linguistic preservation and inclusive digital access.
Khasi belongs to the Khasic branch of the Austroasiatic language family and is spoken by over 1.4 million people in Northeast India. Despite its active use, Khasi remains underrepresented in digital infrastructure and linguistic research.
KhasiBERT is publicly available at:
https://mwirelabs.com/models/khasibert

Artifacts include:
- Model weights and training logs
- Corpus preparation methodology

This resource is intended for researchers, educators, and civic technologists working on low-resource NLP, linguistic preservation, and inclusive AI.

Linguistic Field(s): Computational Linguistics
Language Documentation
Text/Corpus Linguistics

Subject Language(s): Khasi (kha)