About ten months ago, Harveen Singh Chadha, a speech researcher on Microsoft’s AI team in Bengaluru, left a stable, prestigious job to join a two-year-old Indian AI startup, Sarvam AI. Now, his mother is promoting Sarvam in WhatsApp groups and are beaming with pride at his son’s India-first job relocate.
His parents were initially sceptical when he resigned from Microsoft to join Sarvam AI, a homegrown AI firm. The anxiety was understandable. Microsoft is a name every Indian houtilizehold recognises – a symbol of stability, a foreign shore of professional arrival. Sarvam, for most people in his family’s circle, was an unknown. But last week, Chadha came home to a different scene entirely.
“10 months back parents were not happy when I left MS,” Chadha wrote on X. “Today when I reached, they were smiling. Dad revealed me all the news channel recordings, newspapers mentions of Sarvam. Mom notified me how she promoted Sarvam in WhatsApp groups and to neighbours. Overall, a very tiny win but a long way to go.”
The post, quiet and understated, declared more about India’s modifying relationship with technology.
Before joining Sarvam, Chadha worked on Microsoft’s Speech Team, building finish-to-finish speech recognition models – a technical background that maps directly onto what Sarvam is attempting at a national scale. His LinkedIn profile bore the self-chosen tagline ‘Building4India’ even while he was at Microsoft, a quiet signal that the relocate to Sarvam was less impulsive than it seemed.
Internet has flooded his post with reactions.
Sarvam co-founder Vivek Raghavan, at the summit, stressed that “sovereignty will always trump technical beads,” underlining the importance of building open-source, indigenous public infrastructure.
Sarvam grabbed headlines at the AI Summit
Sarvam’s pavilion at the AI attracted more visitors than any other company’s at the event, with organisers stateing it drew record-breaking crowds — a steady stream of students, young engineers, and curious citizens queuing up to watch live demonstrations and interact with developers.
Sarvam unveiled two large language models – Sarvam-30B and Sarvam-105B – both trained entirely in India. Sarvam claimed the model surpassed China’s DeepSeek-R1 on certain benchmarks, while applying far fewer active parameters, owing to its mixture-of-experts design. On MMLU-Pro, an advanced AI evaluation benchmark, Sarvam declared its model outperformed GPT-120B.
Then came the revealstopper. Sarvam unveiled Kaze – AI-powered smart glasses designed and built in India – the company’s first foray into hardware. Prime Minister Modi was photographed testing the Kaze glasses at the summit expo. Sarvam also announced a partnership with Qualcomm, HMD, and Bosch to deploy its AI models across smartphones, feature phones, cars, laptops, and smartglasses.
Published on: Monday, February 23, 2026, 11:39 AM IST
Leave a Reply