Multilingual Understanding for Northeast India

NE-BERT is a foundational language understanding model created by MWire Labs to support the rich linguistic diversity of Northeast India. Built on the ModernBERT architecture, NE-BERT is trained on curated regional corpora spanning Khasi, Garo, Pnar, Assamese, Manipuri (Meitei), Mizo, Nyishi, Kokborok and Nagamese, along with English. The model addresses long-standing challenges in Northeast NLP: orthographic variation, dialect shifts, spelling inconsistencies, and highly code-mixed text. NE-BERT provides strong performance in classification, NER, retrieval, embeddings, and contextual sentence understanding. It is built to serve real-world needs in local governance, citizen services, rural institutions, educational platforms, and community-driven digital applications. Developers can use NE-BERT to build multilingual search engines, automated classification systems, content moderation tools, civic data analyzers, and educational assistants. Its robustness to irregular text makes it suitable for field reports, social media scraping, district-level documents, and unstructured datasets collected across the region. As part of a long-term effort to strengthen AI research in the Northeast, NE-BERT is openly released for public use through Aikosh and HuggingFace.