In 2010, a software engineer started work on the largest identity system anyone had tried to build, and by one profile he did it for free. Vivek Raghavan joined the Aadhaar programme as its chief biometric architect in a voluntary capacity. No salary was the point. The problem was.
Most of what the public knows about him begins there, which is itself a finding. Where he grew up, what his parents did, what he feared at twenty: none of it turns up in the sources. His education is reported two ways, a PhD from Carnegie Mellon in one bio and an IIT Delhi degree in another, and neither could be pinned to a primary record. The honest version of this story has a hole where the childhood should be.
Each job had been public-interest work, often done without the usual startup rewards.
What the record does show is a long, unglamorous apprenticeship in building things for a billion people. After Aadhaar he worked in product at UIDAI, helped bring AI into identity checks and fraud detection, then moved to the EkStep Foundation as its chief AI evangelist. At AI4Bharat, the research lab at IIT Madras, he mentored alongside Pratyush Kumar. The two were working on open-source AI for Indian languages while the world was still arguing about whether chatbots could be trusted.
What almost nobody noticed was the pattern. Each job had been public-interest work, often done without the usual startup rewards. In August 2023 the two men turned that habit into a company. Sarvam AI raised roughly $41 million that December from Lightspeed, with Peak XV and Khosla Ventures joining. That was a large cheque for an Indian AI seed, and a small one against the labs in San Francisco.
The bet was narrow on purpose. Sarvam would build for Indian languages and for voice, not clone an English chatbot. It released Sarvam-1 in October 2024. In April 2025 the IT ministry picked it to build an indigenous foundation model under the IndiaAI Mission, with government-backed GPU compute, and UIDAI announced work on voice features for Aadhaar services. Then came the heavier models, 30B and 105B, shown at the India AI Impact Summit in February 2026, and in March the weights went out openly under an Apache 2.0 licence.
There is a cost to that kind of ambition, and the sources only hint at it: public money brings public scrutiny, and the company has not disclosed revenue. In June 2026 it raised $234 million at a $1.5 billion valuation, with HCLTech putting in $150 million, which made it a unicorn.
For a first-time founder the lesson sits in the early years rather than the valuation. Raghavan spent over a decade on problems that paid in credibility, not cash, and when he finally asked investors for money, he already knew the problem better than anyone in the room. The unpaid years, it turns out, were the research.