IIT-M-incubated Bodhan AI unveils open models for Indian languages

CHENNAI: Bodhan AI, a Centre of Excellence in AI for Education incubated at IIT Madras, on Friday unveiled a suite of open foundational AI models for Indian languages in collaboration with NVIDIA, aimed at expanding access to multilingual technology for education and public-interest applications.

Developed with AI4Bharat, the models cover four capabilities—speech recognition (Indic-Transcribe), text-to-speech (Indic-Speak), machine translation (Indic-Translate) and optical character recognition (Indic-OCR).

The initiative will enable developers, startups, universities, researchers, technology companies and government agencies to deploy, adapt and fine-tune the models for Indian-language applications, including those involving regional dialects and accents. Bodhan AI and AI4Bharat are using NVIDIA’s NeMo framework to train models for automatic speech recognition, machine translation and optical character recognition.

NVIDIA Nemotron 3.5 ASR has also been post-trained for Indian languages, while NVIDIA TensorRT-LLM and vLLM inference microservices are being used to serve them. “Bodhan AI and AI4Bharat aim to give India’s developers and model builders the best open models they can deploy, adapt and fine-tune,” said Prof Mitesh Khapra, principal investigator at Bodhan AI and AI4Bharat.

NVIDIA senior distinguished engineer Niket Agarwal said, “Open models are essential for developers building technology for the languages and communities they serve. The models will be offered as open-weight releases and through hosted APIs.”

Bodhan AI said that its educational applications would remain free for learners, educators and partner state governments. Its API infrastructure is also intended to operate within India’s digital ecosystem, enabling large-scale deployment and customisation for Indian languages and contexts.