Vivek Raghavan transitioned from deep-tech engineering to national-scale digital infrastructure, contributing to Aadhaar and shaping India’s technology-first governance approach.
Sarvam AI focuses on Indic language models, enabling inclusive AI access across diverse linguistic populations often excluded from English-dominated systems.
His vision emphasizes sovereign AI infrastructure, ensuring India builds, controls, and scales its own intelligence systems amid rising global competition.
On a humid Bengaluru evening, in a room full of engineers debating model benchmarks and latency curves, Vivek Raghavan pauses not to correct a number, but to reframe the question. “We’re not building for leaderboard scores,” he says quietly. “We’re building for people who’ve never been part of the conversation.” The room resets. That instinct to zoom out from metrics to meaning has defined much of his career.
Unlike many other AI founders who ride trends, Raghavan is no stranger to the deeper technology and infrastructure of systems. Having been trained at both IIT Delhi and Carnegie Mellon University, his early career was rooted in electronic design automation an industry that works way behind the scenes of most consumer tech. It was painstaking work—designing tools that design chips but it also inculcated a philosophy: build systems that others depend upon.
This philosophy reached its peak when he became an important part of India’s Aadhaar project. Working on what is one of the biggest digital identification systems in the world, Raghavan experienced the challenges associated with designing for scale – not millions but billions of interactions taking place in a variety of languages, geographies, and socio-economic settings. In this scenario, technology did not merely mean software; it meant infrastructure.
In years to come, as generative AI started redefining the global technology discourse, Raghavan would see a similar scenario unfolding. This time, however, the infrastructure in question was not one of physical technology but that of intelligence. And once again, India could become a user of someone else’s system. This understanding prompted the creation of Sarvam AI in 2023, along with co-founder Pratyush Kumar. The company’s mission is simple yet ambitious: to create AI models that comprehend India, not in addition to something else, but for it. Raghavan says, “Language is the interface to intelligence. If AI doesn’t work in your language, it doesn’t really work for you.”
The company Sarvam AI’s mission revolves around building Indic language models that can cater to the diverse linguistic environment in India. While most AI systems developed across the globe do well in English, Sarvam specializes in creating models for languages like Hindi, Tamil, Bengali, and many others, including dialects that don’t get featured in digital language corpora.
It’s not just a matter of translation; Sarvam focuses on building understanding and interpreting culture, idioms, and context. This vision of theirs is driven by a larger idea – one that suggests that AI needs to be accessible to everyone to have any meaning. In a country where hundreds of millions of people use technology in languages other than English, an AI-first strategy is counterproductive.
However, Raghavan’s ideas go beyond Indic languages. He often mentions the concept of sovereign AI, which refers to the notion that every nation-state should have its own AI system that is controlled by itself.
“We’ve seen this before with other layers of technology,” he notes. “If you don’t build it yourself, you end up adapting to someone else’s priorities.” India, in his view, has a unique opportunity. It involves scale, digital public infrastructure, and an increasingly robust community of developers. The emergence of generative AI has further strengthened its progress. Startups are innovating, enterprises are adopting, and the government is pouring money into projects such as the IndiaAI Mission.
But this is easier said than done. There are structural limitations faced by Indian AI startups. They lack access to quality data and computing power and are competing against deep-pocketed global companies. There is regulation, too, which is a must, although nascent. But Raghavan does not shy away from these limitations; he factors them in. “Constraints force clarity,” he says. “We can’t outspend global companies, so we have to out-focus them.”
This ideology is reflected in Sarvam’s products, hiring approach, and collaborations. The company is guided by a systems approach—combining research, infrastructure, and applications in one rather than separating them out as distinct entities. His peers say that Raghavan is calm, thoughtful, and always paying attention to the fundamentals. He shies away from grand statements in favor of sharp questions: Who is it for? What problem is it solving? Can it be scaled responsibly?
His long-term vision reflects his earlier research on digital public infrastructure, where the technology was leveraged to enable identity at scale. Similarly, Raghavan believes that there is a possibility of using AI to enable intelligence at scale provided this is achieved in a responsible manner. This involves finding the right balance between innovation and responsibility, speed and inclusivity. India is very much part of the global AI race, but its standing is not yet fixed. It has everything needed to emerge as a leader in certain spheres, especially those that involve large-scale multilingual deployments. But leadership is not something that happens by itself.
Back in that Bengaluru office, the conversation eventually returns to benchmarks and models. The perspective, however, has changed. For Raghavan, the issue is never about the extent to which the technology might evolve. Rather, it is about who might get to wield it, and under what conditions. For in Raghavan’s calculations, the future of artificial intelligence in India will not depend upon any particular innovation, but rather on an even more profound challenge: Will a billion people ever have a voice in their own thinking?