Overview
- Bodhan AI, the IIT Madras‑incubated Centre of Excellence, publicly released four foundational models for automatic speech recognition, text‑to‑speech, optical character recognition and machine translation after announcements on Friday and Saturday.
- The models carry open weights and are offered via hosted APIs on sovereign digital infrastructure so institutions can run or call the models without sending raw data to third‑party servers.
- Language coverage varies by capability with ASR supporting 27 Indian languages, OCR 23, TTS 23 and translation 22, enabling voice input, speech output, printed and handwritten text reading, and inter‑language conversion.
- Bodhan AI also rolled out a Student Tutor Bot for Classes 6–12 and a Teacher Assistant Bot aligned to NCERT and SCERT curricula; teachers can review, edit, regenerate or discard any AI output to retain classroom control.
- The project was developed with AI4Bharat and uses post‑training on NVIDIA NeMo/Nemotron models, and the teams say the open releases aim to reduce duplicated engineering effort and speed localised EduAI tools for India’s many languages.