Northeast India · AI Deployment Lab · Shillong, Meghalaya

AI that works
where it matters.

We build and deploy language AI for Northeast India’s 220+ indigenous languages, from government services to offline community tools.

Language pairs
0 +
Open models
0 +
Target Langauges
0 +
Open datasets
0 +
Production-ready · Offline-capable · CERT-In aligned · DPDP Act compliant · GIGW 3.0 ready

Lemka · Voice AI

Speech intelligence for Northeast India

Speech-to-text and text-to-speech for Northeast Indian languages, built to run securely on your own infrastructure.

STT + TTS across 7 languages

Khasi, Garo, Mizo, Kokborok, Assamese, Nagamese, Meitei

Speech-to-Text

Real-time transcription across all 7 languages, tuned for regional accents, code-switching, and low-bandwidth environments.

Text-to-Speech

Natural-sounding voice generation across the same 7 languages, running fully on your infrastructure, no data leaves the deployment.

Licensing

  • Software license, per deployment or on-prem
  • Implementation and setup
  • Annual maintenance contracts
  • Language and dialect optimization

Who we work with

01

Government & Public Sector

Multilingual citizen services, document processing, and offline-capable public tools — compliant with Indian data governance standards.

Explore Government Solutions →

02

Enterprise & Business

Localized deployment and runtime licensing for regional businesses, from BFSI to logistics, retail, and beyond

Start Building →

03

Developers & researchers

Open datasets, pretrained models, and co-authorship opportunities at ACL, EACL, EMNLP. All models released under CC-BY-4.0.

View publications & datasets →

Real deployment.
Not just research output.

Every model we build is designed for Northeast India’s real conditions — low connectivity, regional scripts, and multilingual contexts that global AI ignores.

Offline-capable models for areas with limited internet access

Government document processing in Khasi, Garo, Mizo, Kokborok and Meitei

Voice interfaces that work in local languages, not transliterated English

Built in the open

Open source. For everyone.

All foundational models are freely available on HuggingFace and listed on AIKosh, India’s National AI Repository.

Free to download and use: research, education, and product development, no restrictions

Peer-reviewed research: open source models backed by a published or submitted paper at ACL, EACL, EMNLP

Community datasets collected with consent, annotated locally, and shared openly

The NE-Stack

NE-ASR

8 languages · Whisper-medium fine-tuned

NE-OCR

9 scripts · 95.51% accuracy · ViTSTR-Base unified model

Kren – M

Bilingual (Khasi–English), Generative Language Model. ~2.6B params

NE-BERT

149M params · 9 NE languages · ModernBERT

Also open-source: NE-Trans · NE-CLIP · NE-LID · NE-MultiSpeech · NE-SpeechEmbed

Ready to deploy AI
in your language?

Government departments, NGOs, developers, researchers; we work with all of them. Let’s talk about what you’re building.