India gave the world the zero. There is no reason it cannot give the world its next foundation models — and every reason it should build them in its own languages, on its own terms.
For most of the AI era, we have been consumers of intelligence built elsewhere. The models that increasingly mediate knowledge, commerce and public life were trained on other languages, aligned to other contexts, and governed under other laws. That is not a criticism of those who built them. It is a description of a gap — and gaps this fundamental do not close themselves.
That is why Tanumanasa builds foundation models for Bharat: models where Hindi, Odia, Bengali, Tamil and the languages of a billion people are first-class citizens, not translation targets. Where the idioms, knowledge systems and realities of India shape the training data rather than appearing in it by accident.
We build from Odisha deliberately. Kendujhar and Bhubaneswar are not the periphery of the technology world — they are proof that the next generation of deep-tech can begin anywhere there is conviction, discipline and talent. The state that carved the Konark wheel can build the machinery of intelligence.
The Indian way in AI, as I see it, is efficient, responsible and plural: models that are affordable at Indian scale, evaluated honestly, deployed with care, and informed by our own intellectual traditions — Tanu, the form; Manasa, the mind. Giving form to intelligence, and mind to machines.
This will take a decade, many partners, and a generation of researchers who choose to build here. If any part of that sounds like you — a policymaker, a scientist, an engineer, a student in a district school asking questions in her own language — this vision has room for you in it.
Building Bharat's next foundation models — from Odisha, for India, for the world.