Bodhan AI and AI4Bharat Launch Open Indic AI Models with NVIDIA Nemotron to Advance Multilingual AI in India
-
Bodhan AI and AI4Bharat have announced a suite of open AI models for Indian languages in collaboration with NVIDIA, targeting multilingual voice, translation, vision and educational applications across India.
Chennai, September 11, 2026: Bodhan AI, a Centre of Excellence in AI for Education incubated at IIT Madras, has announced a suite of open, state-of-the-art foundational AI models for Indic languages, developed in collaboration with NVIDIA. The initiative is aimed at strengthening India's open AI ecosystem and expanding access to technologies that can process and generate content across the country's diverse regional languages.
The initiative brings together Bodhan AI, AI4Bharat and NVIDIA to develop and make available AI capabilities spanning speech recognition, text-to-speech, machine translation and optical character recognition. The models are intended to support developers, researchers, educational institutions, startups, technology companies and government organisations building applications for Indian-language users.
AI4Bharat, based at IIT Madras, contributes expertise in Indic-language modelling, multilingual technologies and datasets. Together, the organisations are working to make advanced Indian-language AI models available for adaptation, deployment and fine-tuning.
Four core Indic AI capabilities
The models announced under the initiative cover four major areas of language and document intelligence:
- Indic-Transcribe – speech recognition for Indian languages.
- Indic-Speak – text-to-speech capabilities.
- Indic-Translate – machine translation across Indian languages.
- Indic-OCR – optical character recognition for extracting information from documents and images.
The initiative is particularly focused on applications in education, where voice, text and scanned-document technologies can help make learning resources more accessible to students and educators who primarily use regional languages.
The organisations said the broader objective is to address the continuing gap between the large number of Indians who interact with technology in regional languages and the availability of reliable AI models capable of supporting those languages.
NVIDIA technology to support training and deployment
Bodhan AI and AI4Bharat are using the NVIDIA NeMo framework to train models for automatic speech recognition, machine translation and optical character recognition.
The collaboration also includes post-training NVIDIA Nemotron 3.5 ASR to support Indian languages, including regional dialects and accents. For inference and deployment, the models are served using NVIDIA TensorRT LLM and vLLM inference microservices.
According to the organisations, NVIDIA and the AI teams are also collaborating on datasets, training approaches and evaluation methodologies that can contribute to future foundational models for Indian languages.
The use of these technologies is intended to help organisations train, customise and deploy multilingual AI systems more efficiently.
Open models aimed at India's developer ecosystem
A key aspect of the initiative is its emphasis on openness. Bodhan AI and AI4Bharat said the models will be made available through open-weight releases as well as hosted APIs.
This approach gives researchers, startups and technology companies the ability to experiment with and fine-tune the models for specialised applications. Organisations that require faster deployment can use hosted APIs to integrate the capabilities into their own products and services.
Bodhan AI said its educational applications will remain free for learners, educators and partnering State Governments.
The initiative could therefore provide a foundation for applications that require Indian-language speech processing, translation, document digitisation and other AI capabilities without requiring organisations to build every underlying model from the beginning.
Focus on AI for education
Education remains a central application area for Bodhan AI's work.
The models are designed to support learners and teachers across voice, text and scanned documents, potentially enabling more personalised and accessible educational experiences in regional languages.
For students, multilingual AI can help bridge language barriers in learning materials and digital educational resources. Teachers and educational institutions can potentially use speech, translation and document-processing capabilities to develop or adapt content for different linguistic audiences.
The broader vision is to make advanced AI capabilities available in contexts where English-language technology may not adequately serve the linguistic needs of Indian learners.
What Prof. Mitesh Khapra said
Prof. Mitesh Khapra, Principal Investigator at Bodhan AI and AI4Bharat, said the collaboration is intended to give Indian developers access to open models that can be deployed, adapted and fine-tuned.
He said the partnership with NVIDIA would help accelerate the availability of open, advanced AI capabilities for the Indian-language ecosystem and support faster innovation and adoption of multilingual and multimodal AI.
NVIDIA highlights role of open AI models
Niket Agarwal, Senior Distinguished Engineer, NVIDIA, said open models are important for enabling developers to build AI systems suited to the languages and communities they serve.
He highlighted NVIDIA accelerated computing, NVIDIA Nemotron and NVIDIA TensorRT as technology foundations for training, customising and deploying the models.
The collaboration reflects the growing focus on developing AI systems that can support India's linguistic diversity while enabling developers and researchers to build applications for specific local requirements.
Building an AI ecosystem for Indian languages
The initiative goes beyond individual AI models by seeking to create a broader ecosystem around Indian-language technology.
The open-weight models can be used by researchers and developers for experimentation and specialised applications, while APIs can provide organisations with an easier route to deployment.
Potential users include:
- Ed-tech companies
- Universities and educational institutions
- AI researchers
- Startups
- Technology enterprises
- Government agencies
- Developers working on public-interest applications
By making language technologies more accessible, the initiative aims to encourage more organisations to develop applications that work effectively across India's regional languages.
AI infrastructure and India's sovereign digital ecosystem
Bodhan AI also highlighted sovereign digital public infrastructure for AI in education as a core part of its approach.
The organisation said such infrastructure is important for operating AI systems at population scale and for customising and managing models according to India's languages and requirements.
Its API infrastructure is intended to be hosted within India's sovereign digital ecosystem, supporting its stated vision of "AI for India, governed in India."
This focus places infrastructure, language accessibility and local deployment alongside the development of AI models themselves.
Why Indic-language AI matters
India's multilingual environment creates a distinct challenge for artificial intelligence. A large section of the population communicates, studies and accesses information in languages other than English.
AI systems that can accurately understand speech, translate languages, read documents and generate spoken content can therefore play an important role in expanding digital access.
The Bodhan AI-AI4Bharat initiative combines these capabilities into a broader platform for Indic-language applications. Its focus on open models also provides opportunities for developers to adapt the technology to specific sectors and use cases.
For education in particular, multilingual AI could help institutions create learning experiences that are better aligned with the linguistic backgrounds of students.
The road ahead
Bodhan AI, AI4Bharat and NVIDIA's collaboration represents an effort to strengthen India's indigenous and open AI capabilities through a combination of research, datasets, computing infrastructure and deployment technologies.
With applications spanning speech, translation, text-to-speech and document understanding, the models have potential relevance across education and other public-interest sectors.
The initiative's long-term significance will depend on the continued improvement of model accuracy, language coverage, accessibility, responsible deployment and the ability of developers and institutions to adapt the technology to real-world requirements.
For now, the release provides India's AI developer and research ecosystem with another set of tools focused specifically on the country's diverse linguistic environment.
Key Facts
| Particular | Details |
|---|---|
| Organisation | Bodhan AI |
| Collaboration | AI4Bharat and NVIDIA |
| Location | Chennai, Tamil Nadu |
| Focus | Indic-language and multilingual AI |
| Key Models | Indic-Transcribe, Indic-Speak, Indic-Translate, Indic-OCR |
| Major Application | Education and public-interest applications |
| AI Framework | NVIDIA NeMo |
| Speech Technology | NVIDIA Nemotron 3.5 ASR |
| Inference Technologies | NVIDIA TensorRT LLM and vLLM |
| Availability | Open-weight releases and hosted APIs |
| Educational Applications | Free for learners, educators and partnering State Governments |
Click Here for More Science & Technology