Bodhan AI, a Centre of Excellence in AI for Education incubated at the Indian Institute of Technology (IIT) Madras and funded by the Union Ministry of Education, has rolled out four foundational artificial intelligence models aimed at bolstering regional language capabilities across India. Developed alongside AI4Bharat, the suite covers automatic speech recognition, speech generation, machine translation, and optical character recognition to establish a sovereign technology backbone for digital learning ecosystems.
The models have been made available to the wider technology and educational sector as digital public goods through open-weight releases and hosted application programming interfaces. Supported by sovereign digital infrastructure, the initiative is designed to prevent fragmented development by providing a common layer that edtech companies, startups, researchers, and government institutions can build upon without duplicating foundational research efforts.
Core Capabilites and Linguistic Reach Across 27 Languages
The newly launched model suite addresses diverse linguistic requirements, providing varying levels of support across 27 Indian languages and regional dialects. The automatic speech recognition model leads the coverage by supporting 27 languages, followed by optical character recognition at 23 languages, text-to-speech generation at 23 languages, and Bodhan-Translate at 22 languages. According to project disclosures, the automatic speech recognition system was post-trained specifically to accommodate regional accents and diverse speech patterns native to different parts of the country.
Technological optimization relies heavily on advanced infrastructure frameworks. The foundation models were trained and refined using NVIDIA Nemotron open models and the NVIDIA NeMo framework for automatic speech recognition, machine translation, and optical character recognition. Furthermore, the services are delivered using NVIDIA TensorRT-LLM and vLLM inference microservices. Explaining the broader vision for the ecosystem, IIT Madras Director V Kamakoti stated, "India’s AI journey cannot be built on technology alone. It must be built on technology that understands India. The launch of these voice and vision models is an important step towards creating sovereign Digital Public Infrastructure for AI in education that can serve our linguistic diversity and enable innovation at scale."
Powering Dedicated Student and Teacher Applications
Beyond serving as an underlying toolkit for external developers, the foundational models directly power two proprietary applications developed by Bodhan AI: the Student TutorBot and the Teacher Assistant Bot. The Student TutorBot functions as a personalized learning companion for students in Classes 6 through 12, structured around National Council of Educational Research and Training and State Council of Educational Research and Training curricula. Learners can interact with the system via text or voice across 22 Indian languages, drawing directly from official textbook content to receive step-by-step explanations, diagrams, calculations, and assessments.
In parallel, the Teacher Assistant Bot serves as an administrative workspace designed to assist educators in planning and executing classroom work. Instructors can specify parameters such as grade, subject, topic, duration, and difficulty to generate tailored lesson plans, quizzes, homework assignments, and revision sheets. Teachers can also upload handwritten or scanned student work to evaluate it against specific marking criteria. Project leadership emphasized that AI outputs are treated strictly as starting points rather than autonomous decisions, allowing teachers to review, edit, regenerate, or discard generated content.
Mitigating Ecosystem Fragmentation Through Open Infrastructure
Principal Investigator Mitesh Khapra emphasized that the initiative is structured to collaborate with the existing technology landscape rather than compete against it. "Bodhan AI aims to build with the ecosystem, not compete with it. 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," Khapra explained. By pooling resources into a shared foundational layer, smaller enterprises and universities gain access to advanced vision and speech capabilities that would otherwise require prohibitive computational investments.
This open-weight availability enables edtech developers to integrate voice navigation, localized translation, and document comprehension into existing platforms via APIs. Startups can fine-tune the weights for niche applications, while academic institutions can utilize the codebase for advanced linguistic research. The shared architecture effectively lowers the barrier of entry for localized digital tools, ensuring that smaller regional educational providers can deploy sophisticated language models without building complex training pipelines from scratch.
Data Governance and Sovereign Deployment Frameworks
Compliance, data privacy, and security remain central to the deployment strategy. Bodhan AI has integrated strict data anonymisation protocols into its system architecture to align with national education data frameworks. The hosted API infrastructure operates on a sovereign deployment model, granting participating institutions and government partners strict control over how data is handled and processed.
This focus on data governance is intended to reassure educational authorities concerned about student privacy and data sovereignty. By ensuring that localized interactions and educational datasets remain governed by national standards, the initiative establishes a secure pathway for integrating artificial intelligence into mainstream public and private schooling systems across India without compromising sensitive user information.
