IIT Madras' Bodhan AI launches four AI models for Indian languages in partnership with AI4Bharat

IIT Madras-incubated Bodhan AI on Friday launched four foundational AI models for Indian languages in partnership with AI4Bharat as it looks to build a common technology layer for education-focused AI applications in the country.

Developed in partnership with IIT Madras' AI4Bharat research lab, the models cover speech recognition, speech generation, machine translation and optical character recognition (OCR). Bodhan AI is making them available as open-weight models as well as hosted APIs on sovereign digital infrastructure, allowing startups, edtech companies, researchers and government institutions to build and customise applications on top of them.

The models are part of the Bharat EduAI Stack, which Bodhan AI is positioning as sovereign digital public infrastructure for India's multilingual education ecosystem.


Also Read: India tops global AI adoption with highest share of ‘frontier professionals’: Microsoft

The broader aim is, according to the company, to avoid every company, institution or government body having to build the same AI capabilities independently, particularly for India's large and diverse set of languages.

“Bodhan AI aims to build with the ecosystem, not compete with it,” said Prof. Mitesh Khapra, Principal Investigator at Bodhan AI.

“We built these voice and vision models, and made them accessible as Digital Public Goods on a Digital Public Infrastructure, so that efforts across the country don't remain fragmented. Instead, there is one common layer that can power all Edu AI in India, without every institution having to duplicate the same work,” he said.

The models have been trained and optimised using NVIDIA's Nemotron open models and libraries, including the NVIDIA NeMo framework for automatic speech recognition, machine translation and OCR. Bodhan AI said it post-trained NVIDIA Nemotron 3.5 ASR to support Indian languages, including regional dialects and accents.

The models are being served using NVIDIA TensorRT-LLM and vLLM inference microservices. NVIDIA and Bodhan AI are also collaborating on datasets, training recipes and evaluations for future foundational models for Indian languages.

The applications that can be built using these models range from voice-enabled AI tutors to tools that can read and understand educational material in different Indian languages.

For instance, speech recognition can allow students to speak to an AI tutor in their preferred language instead of typing in English. Text-to-speech can allow AI systems to respond in Indian languages, while OCR can help education applications understand textbooks, worksheets and handwritten answers. Machine translation can be used to move educational content across Indian languages.

Bodhan AI is also launching two applications built on these models — Student Tutor Bot and Teacher Assistant Bot.

The Student Tutor Bot is aimed at students in Classes 6–12 and is built around NCERT and SCERT curricula. Students can interact with it through text or voice across 22 Indian languages, with the system using textbook content to provide explanations, examples and assessments.

Also Read: Zoho launches AI-powered Classes 2.0, offers platform free to government schools, colleges and universities in India

The Teacher Assistant Bot, meanwhile, is designed to help teachers create lesson plans, worksheets, quizzes, homework and revision material. Teachers can specify parameters such as grade, subject, topic and difficulty, and upload student work for evaluation against specific marking criteria.

Bodhan AI said teachers will remain in control of the process, with AI-generated content serving as a starting point that can be reviewed, edited, regenerated or discarded rather than being used to make automatic classroom decisions.

Prof. V. Kamakoti, Director, IIT Madras, said India's AI push needs to focus on technology that can understand the country's linguistic diversity.

“India’s AI journey cannot be built on technology alone. It must be built on technology that understands India,” he said.

Bodhan AI said the infrastructure is also being designed around data privacy and sovereign deployment, with data anonymisation protocols and compliance with applicable national education data frameworks.

The larger bet is to create an AI infrastructure layer that Indian edtech companies, startups, researchers and public institutions can use without having to build foundational multilingual AI capabilities themselves.