IIT Madras Launches Bodhan AI Suite with NVIDIA for Indic Languages

IIT Madras-incubated Bodhan AI, a Centre of Excellence focused on Artificial Intelligence for Education, has announced a new suite of open foundational AI models designed for Indian languages. Developed in collaboration with NVIDIA and AI4Bharat, the initiative aims to strengthen multilingual and multimodal artificial intelligence for education and other digital applications in India.

The new suite brings together four major AI capabilities: speech recognition, text-to-speech, machine translation and optical character recognition. The models are designed to address the linguistic diversity of India and support the development of educational tools that can work across languages, accents, scripts and different forms of learning content.

Four AI Models for Indian Languages

The Bodhan AI suite includes four core models, each addressing a specific language technology requirement.

• Indic-Transcribe: Converts speech into text and is designed to handle Indian languages, accents, dialects and code-mixed conversations.

• Indic-Speak: Converts written text into natural speech, supporting Indian languages and multilingual content.

• Indic-Translate: Provides machine translation between English and Indian languages, helping make educational and digital content more accessible.

• Indic-OCR: Extracts text and document structure from printed and handwritten material, including tables and other complex content.

Together, these models can support an end-to-end AI ecosystem for multilingual learning, from converting classroom speech into text and translating educational material to reading documents aloud and digitising printed resources. Bodhan AI's platform currently describes Indic-Translate as supporting English and 22 scheduled Indian languages, while Indic-Speak supports 22 Indian languages and English.

NVIDIA Technology Powers the AI Architecture

The models have been developed using NVIDIA's NeMo framework. The collaboration also incorporates NVIDIA Nemotron 3.5 ASR for post-training speech recognition capabilities aimed at improving performance across Indian accents and dialects.

For inference and deployment, NVIDIA TensorRT-LLM and vLLM-based microservices are being used to support scalable AI applications. The technology is intended to help developers deploy multilingual AI models efficiently for education and other large-scale use cases.

Focus on AI-Powered Education

Education remains a central focus of the Bodhan AI initiative. The models can help developers build regional-language learning applications, voice-enabled educational tools, translated study material and systems that can process documents in Indian scripts.

Indic-Transcribe, for example, is designed to recognise diverse Indian speech, while Indic-Speak can help deliver lessons and other educational material through audio. Indic-OCR can assist in converting printed or handwritten educational resources into digital formats, and Indic-Translate can make learning content available across languages.

Bodhan AI and AI4Bharat have also emphasised the broader objective of creating an open ecosystem in which developers can adapt and deploy multilingual AI technologies for Indian requirements.

Open-Weight Models and Hosted APIs

The foundational models are being released as open-weight systems and are also accessible through hosted APIs within India's digital ecosystem. The Bodhan AI API platform provides services for translation, speech recognition, text-to-speech and document OCR.

According to the announcement, educational applications built using these models are expected to remain free for learners, teachers and participating state governments. This could help expand access to AI-powered educational resources beyond English and other widely supported languages.

The launch marks another step towards developing India's own multilingual AI capabilities, with IIT Madras, Bodhan AI, AI4Bharat and NVIDIA combining research, infrastructure and open technologies to support language-inclusive digital education.