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 Indian languages, developed in collaboration with NVIDIA.
The initiative, undertaken with AI4Bharat, aims to expand access to multilingual and multimodal AI capabilities by giving developers, researchers and enterprises models that can be deployed, adapted and fine-tuned for specialised applications.
Four core AI capabilities
The models span four core capabilities, speech recognition through Indic-Transcribe, text-to-speech through Indic-Speak, machine translation through Indic-Translate and optical character recognition through Indic-OCR.
AI4Bharat brings expertise in Indic-language and multilingual modelling and datasets to the initiative, while Bodhan AI is focused on building an open AI ecosystem with applications particularly relevant to education.
The initiative addresses a persistent gap in India’s technology ecosystem, where millions of people interact with technology primarily through regional languages, while developers often have limited access to reliable speech, translation and document AI models.
NVIDIA technology powers model development
Bodhan AI and AI4Bharat are using the NVIDIA NeMo framework to train models for automatic speech recognition (ASR), machine translation and optical character recognition.
The collaboration includes post-training NVIDIA Nemotron 3.5 ASR to support Indian languages, including regional dialects and accents. The models are served using NVIDIA TensorRT-LLM and vLLM inference microservices.
NVIDIA and Bodhan AI are also collaborating on datasets, training recipes and evaluations to support the development of future foundational models for Indian languages.
“Bodhan AI and AI4Bharat aim to give India’s developers and model builders the best open models they can deploy, adapt and fine-tune, driving fast innovation and adoption of multilingual and multimodal AI across the broader Indian language AI ecosystem. Partnering with NVIDIA lets us bring open, state-of-the-art capability to that ecosystem faster than we could alone,” said Bodhan AI and AI4Bharat, Principal Investigator, Prof. Mitesh Khapra.
“AI is becoming foundational infrastructure for every industry, and open models are essential to ensuring developers can build for the languages and communities they serve. NVIDIA accelerated computing, NVIDIA Nemotron, and NVIDIA TensorRT provide the technology foundation to train, customise and deploy these models efficiently. By using these technologies, organisations can expand access to multilingual AI capabilities across India’s developer and research ecosystem,” said NVIDIA, Senior Distinguished Engineer, Niket Agarwal.
Open models and APIs
Bodhan AI and AI4Bharat are making the models available through both open-weight releases and hosted APIs. Researchers, startups and technology companies can fine-tune the models for specialised applications, while organisations seeking faster deployment can integrate the same capabilities through low-cost APIs.
Bodhan AI’s educational applications will remain free for learners, educators and partnering state governments.
The models are designed to support education across voice, text and scanned documents, potentially making personalised learning and educational resources more accessible to students and teachers in regional languages.
Focus on sovereign AI infrastructure
Sovereign digital public infrastructure for AI in education is a core element of Bodhan AI’s approach. The organisation views such infrastructure as important for operating AI at population scale and customising and running models efficiently for India-specific languages and contexts.
Its API infrastructure is intended to be hosted within India’s sovereign digital ecosystem, reinforcing the broader vision of “AI for India, governed in India.”
The initiative is expected to benefit ed-tech companies, universities, researchers, startups, technology enterprises and government agencies, while strengthening the availability of open Indian-language AI models for education and public-interest applications.