By Amitabh Nag
There is a stage in technological development when adoption stops being the difficult part. The difficult part is building the infrastructure that allows technology to work for problems shaped by our own scale, diversity and context.
India has already demonstrated the power of this approach through Digital Public Infrastructure (DPI). The strength of DPI lies not in any single application, but in the common, interoperable platforms on which governments, businesses and developers can build at population scale. The genius of a platform is that it does not attempt to solve one problem once. It creates a foundation on which many problems can be solved by many people. AI now calls for the same platform thinking.
India’s AI journey will remain deeply connected to the global ecosystem. We will learn from global research, work with international institutions and benefit from advances made around the world. At the same time, confidence in AI means building capabilities for challenges that require a distinctly Indian technological foundation. Language is one such challenge.
India’s linguistic diversity creates complexity across scripts, morphology, syntax, pronunciation, accents, dialects, vocabulary and code-switching. Millions of citizens also encounter technology at the intersection of the language divide, literacy divide and digital divide. A service can be digitally available without being genuinely accessible. Language, therefore, is not simply a translation problem. It is an infrastructure challenge.
This is where BHASHINI (BHASHa INterface for India) assumes significance. Launched in 2022 under the National Language Translation Mission (NLTM), BHASHINI seeks to make language technology available as a digital public good. But its significance lies beyond any individual application. BHASHINI is a platform, a common technological layer that brings together language technologies, AI models, datasets, tools and APIs, allowing them to be discovered, accessed, combined and reused across applications and domains.
That distinction is fundamental. An application addresses a particular need. A platform creates the conditions for an entire ecosystem to innovate. A government department developing a citizen service, a startup building a voice interface, a researcher developing a new model or an enterprise creating a sector-specific solution should not have to rebuild the language layer from the ground up. They should be able to build on what already exists. And as they build, they can contribute new models, datasets, tools and applications back to the ecosystem. The platform grows not merely by addition, but through participation.
What makes BHASHINI a platform is not simply the range of technologies it hosts, but the relationships between them. Models, datasets, APIs, tools and applications form different layers of a common ecosystem. Data can improve models; models can be exposed through APIs; APIs can be integrated into applications; and applications can generate new feedback and use cases. This creates a continuous cycle in which the platform becomes more useful as more participants contribute to and build upon it. The value, therefore, lies not only in what BHASHINI provides, but in what others can build because those capabilities are available as shared infrastructure.
At the technical core of this ecosystem is the National Hub for Language Technologies (NHLT), which supports BHASHINI’s population-scale AI infrastructure. The platform hosts 350+ AI models across capabilities such as speech recognition, neural machine translation and text-to-speech. BHASHINI powers 800+ government websites, has enabled more than 9 billion cumulative AI inferences and processes more than 24 million AI inferences daily. But the significance is not the number itself. The significance is that these capabilities can be brought together as building blocks that others can use and build upon.
The BHASHINI platform spans Automatic Speech Recognition, Neural Machine Translation, Text-to-Speech, Optical Character Recognition, Transliteration, Language Identification, Document Translation and Voice-first Conversational AI. Each capability has value on its own. But the real power comes when they work together. A system can hear, recognise, translate, understand, generate and respond. What appears to the citizen as a simple interaction may depend on several sophisticated technologies working seamlessly underneath. That is the promise of a platform: complexity at the infrastructure layer, simplicity at the point of use.
The platform approach is already visible across very different domains. MahaVISTAAR-AI and Bharat-VISTAAR apply multilingual AI to agriculture; Nyaya Setu to legal access; Arogya Vani to healthcare; CPGRAMS-Samadhan Didi, Rail Madad, UIDAI and Digi Yatra to citizen services; and eGramSwaraj, SabhaSaar and Shrutlekh to governance, documentation and communication. These applications serve different purposes and different users. Yet the underlying language capabilities can be shared. That is precisely what infrastructure should do: disappear beneath the application while making the application possible.
One of the clearest demonstrations of this platform approach is voice-first AI. When language capabilities exist as a common platform, interaction need not begin with a keyboard, a form or a complex digital menu. It can begin with speech. A citizen can speak in a familiar language; speech recognition, translation, language understanding and speech generation can work together in the background; and the response can return in a form the citizen can understand. The platform turns technology from something the citizen must learn into something that can adapt to the citizen.
But no platform is stronger than the resources on which it is built. Good AI begins with data. Indian-language AI needs Indian speech, text, terminology and context. It needs representation of accents and dialects, high-quality translation pairs, domain-specific glossaries, benchmarks and continuous evaluation. BhashaDaan brings citizens into this process by enabling contributions of speech, text and translations, helping strengthen the linguistic resources available to the ecosystem. In a platform model, these resources can become shared foundations rather than remaining locked inside individual projects.
Interoperability is what allows that foundation to travel. Through open and interoperable APIs, language capabilities can become reusable building blocks for governments, researchers, developers and enterprises. A platform is not simply a place where technologies are stored. It is a bridge between capability and application. APIs are what allow that capability to move outward to a government service, a startup product, a research project or an enterprise system. Every new participant can therefore extend the reach of what has already been built.
The scale is significant: BHASHINI supports 36+ Indian text languages, 23 Indian voice languages and 35 international languages. But scale should not be measured only in languages, models or applications. The deeper measure of a platform is what it enables others to build. Its success lies in whether a new innovator can enter the ecosystem faster, whether a new service can become multilingual more easily, and whether an improvement in one part of the system can create value across many others.
Sovereignty is another dimension of platform capability. In February 2026, BHASHINI moved to Indian infrastructure, bringing greater control over the environment in which its language technology, data and AI capabilities are developed and deployed. At population scale, control over data, compute and deployment is part of technological capability. Sovereignty is not isolation. It is the ability to make foundational choices and build with confidence on our own terms.
This is where India’s confidence to build our own becomes meaningful. Building our own does not mean rebuilding everything the world has already built. It means knowing where our circumstances demand our own foundations. We can continue to adopt global advances, collaborate with researchers around the world and participate in frontier AI, while building the language technologies, datasets, models, tools and platforms that reflect India’s realities.
The goal is not to build separately from the world. It is to have the capability to build what our circumstances require.
This is the deeper lesson of applying the DPI approach to AI. The power of a platform lies in reuse, interoperability and scale. The value of BHASHINI is therefore not limited to the applications it powers today. Its greater value lies in providing a common foundation on which an entire ecosystem can continue to build.
For India, multilingual AI is becoming one of those foundational capabilities. We do not need to build everything ourselves. We need the confidence and capability to build what cannot be left unfinished.
That is what it means to build our own.
Amitabh Nag is CEO, Digital India BHASHINI Division. BHASHINI is an AI powered language translation platform, bridging literacy, language, and digital divides. It aims to transcend language barriers, ensuring that every citizen can effortlessly access digital services in their own language.
