IIT Madras’ Bodhan AI, AI4Bharat launch open AI models for Indian languages

Bodhan AI, an IIT Madras-incubated Centre of Excellence in AI for Education, announced on Friday a suite of open foundational AI models for Indic-languages developed in collaboration with NVIDIA. The initiative works alongside AI4Bharat to release models spanning four core capabilities: speech recognition (Indic-Transcribe), text-to-speech (Indic-Speak), machine translation (Indic-Translate), and optical character recognition (Indic-OCR), according to a press release.

Also read IIT Madras’ Bodhan AI, AI4Bharat launches four open models for Indian languages

The models were developed using the NVIDIA NeMo framework, which includes post-training NVIDIA Nemotron 3.5 ASR to support Indian regional dialects and accents. Inference is served using NVIDIA TensorRT LLM and vLLM microservices. The initiative aims to accelerate the adoption of multilingual AI solutions across the broader Indian language AI ecosystem, specifically focused on education. Prof Mitesh Khapra, Principal Investigator at Bodhan AI and AI4Bharat, stated that the NVIDIA partnership accelerates bringing open, state-of-the-art capabilities to developers.

These foundational models are accessible through open-weight releases and hosted APIs within India's sovereign digital ecosystem. Educational applications leveraging these models will remain free for learners, teachers, and partnering state governments.