AI4Bharat
AI4Bharat is an IIT Madras research lab that develops datasets, tools and trained models for Indian languages. Its work includes language understanding, translation, speech and transliteration.
Company & leadership
- Origin
- The team's presentation dates the initiative's launch to July 2019.
- Institutional home
- Indian Institute of Technology Madras, Chennai, India; listed by the Wadhwani School of Data Science and AI.
- Research leadership
- Mitesh M. Khapra's current faculty biography says he heads the AI4Bharat Research Lab. This is an academic leadership role, not a company CEO title.
- Organization and support
- A research lab within IIT Madras. The Nilekani Centre was launched in July 2022 with philanthropic support from Rohini and Nandan Nilekani.
- Collaborators
- The lab's research credits students, volunteers and external researchers. Recent Indic models are jointly published with Bodhan AI.
Founding team
- The early project presentation identifies Mitesh M. Khapra, Pratyush Kumar and Anoop Kunchukuttan as the team behind the initiative.
The lab grew from an Indian-language research initiative
The initiative's early mission focused on datasets and models that others could use. The 2022 Nilekani Centre announcement describes foundational language research and the release of tools, datasets and pretrained models for other organizations.
Early models and later adaptations have different origins
AI4Bharat dates IndicBERT to 2020 and IndicBART to 2021. Both were pretrained from the beginning. Later projects also adapt released language models. The distinction matters: an AI4Bharat fine-tune has its own training work, but its base architecture and upstream terms remain part of the model's history.
Model papers identify the contributing teams
IndicBART's release credits Raj Dabre, Himani Shrotriya, Anoop Kunchukuttan, Ratish Puduppully, Mitesh Khapra and Pratyush Kumar. The lab's releases are research collaborations, so individual model cards are more precise attribution records than a single company founder list.
Joint Bodhan releases are listed under their publisher
The September 2026 Indic-Transcribe announcement names Bodhan AI and AI4Bharat together. The Bodhan AI profile contains the joint Indic-Transcribe, Indic-Speak, Indic Translate and IndicOCR articles. They are not counted again here.
The catalog includes usable releases and research artifacts
The official model account also contains older experimental repositories, sparse cards and alternate packages. This guide groups packaging changes and training sweeps where appropriate. A repository name alone does not establish a current API, complete license terms or a separate trained model.
Some published repositories need clearer model documentation
Sparse or template cards remain for IndicGPT, gemma-hi-en-cpt, IndicTrans3-llama-beta and the repository named IndicConformer. This review does not infer their training history or equate them with better documented releases.
Older IndicBERT alpha, Wiki, TLM, large and task-head repositories also lack enough release-specific documentation to establish all relationships. Six early VITS experiment repositories have similarly limited cards. They are not counted as fully reviewed model articles here.