How two engineers built Sarvam AI from an idea to a summit showcase

The summit spotlight was recent, but Sarvam AI's rise began years earlier, as two engineers built an India-first AI startup focused on sovereign models, local datasets and real-world applications

From an idea sketched between two engineers to a presence at India’s biggest AI summit, Sarvam AI’s journey unfolded at speed.

 

At the heart of Bharat Mandapam, where India is hosting the first-ever AI Impact Summit, one startup stood out. Sarvam AI, not yet three years old, found itself at the centre of attention as its made-in-India large language models (LLMs) and new hardware offerings entered the national spotlight.

 

The moment that travelled far beyond the venue came when Prime Minister Narendra Modi was seen wearing a sleek, black pair of AI-driven spectacles at the India AI Impact Summit 2026. The wearable, called Sarvam Kaze, had been developed by Sarvam AI.

 

 

Yet it was not merely the optics that drew attention. The underlying technology prompted a reassessment among some in the tech community. Entrepreneur and technologist Deedy Das admitted that he had initially questioned Sarvam’s focus on smaller Indic language models, but later reconsidered.

 

“When I wrote about them a year ago, I felt like the direction to train small ‘Indic’ language models was wrong. But they have turned it around. They have the best text-to-speech, speech-to-text, and OCR models for Indic languages, and that is actually valuable,” he posted on X, adding that the pricing was reasonable and the products addressed gaps larger global labs were unlikely to prioritise.

 

The summit moment was striking. But the journey that led to it had begun much earlier.

Building sovereign AI for India

Founded in 2023 and headquartered in Bengaluru, Sarvam AI was among 12 organisations tasked by the Indian government with developing AI models built on Indian datasets. The company focused on language and multimodal AI systems tailored to Indian use cases. Its work spanned document processing, speech recognition and language understanding, with training data drawn from Indian languages, scripts and real-world material including documents, textbooks, newspapers and scanned records.

 

The company’s stated aim — “AI for all from India” — was rooted in the idea of sovereign AI. The founders sought to build generative AI systems that understood India’s languages, culture and context rather than relying solely on models trained primarily on Western data.

 

Although Sarvam AI itself was new, the thinking behind it had been shaped over decades.

The brains behind Sarvam AI

Pratyush Kumar and Vivek Raghavan were not first-time techies experimenting with a trend. Both studied Electrical and Electronics Engineering at the Indian Institutes of Technology; Kumar at IIT Bombay and Raghavan at IIT Delhi.

 

Kumar went on to earn a PhD in Computer Engineering from ETH Zurich in 2014. He joined IBM Research in 2015 as a research scientist, working on advanced AI and computing projects, and later spent two years at Microsoft Research. Between 2018 and 2021, he served as faculty at IIT Madras and continued to remain associated as adjunct faculty.

 

Raghavan’s path was different. In a 2024 interview with Forbes, he reflected that “life had its own plans”. After completing his Master’s and PhD at Carnegie Mellon University, he immediately started a company rather than pursuing a conventional academic or corporate track.

 

“That was not the plan, but that is how it worked out,” he said, describing how his first venture operated in electronic design automation (EDA) and was later acquired. His second startup, Mojave, also worked in EDA. Nvidia had been among its early customers. Over nearly two decades, he worked in software that enabled chip design and manufacturability.

 

In 2007, following the passing of his mother, Raghavan returned to India. That return marked a shift towards public digital infrastructure and language technology.

 

He served as chief product manager and biometric architect at the Unique Identification Authority of India (UIDAI), where he worked on the design and scaling of the Aadhaar technology platform. He joined when the first Aadhaar number was issued and remained associated through the milestone of the billionth enrolment. He introduced AI systems to improve service quality and detect identity fraud.

 

Raghavan was also chief AI evangelist at EkStep Foundation and advised Digital India Bhashini under the National Language Translation Mission. He served as chief mentor at the Nilekani Center at AI4Bharat, IIT Madras, which aimed to achieve English-equivalent language AI capabilities for Indian languages through open datasets and models. He contributed to AI models for fraud detection in GSTN, advised the National Payments Corporation of India and contributed to the Data Empowerment and Protection Architecture (DEPA) framework. He was part of the Supreme Court’s AI committee and oversaw the rollout of SUVAS for translating judgments into Indian languages. Earlier, he had also volunteered as CTO for Team Indus, India’s entry to the Google Lunar X Prize.

 

The emergence of generative AI globally, particularly after the release of ChatGPT by OpenAI, created what Raghavan saw as an opening. In late 2023, soon after Sarvam AI raised its first major funding round, he outlined a vision of building foundational models rooted in Indian data and realities.

Funding, partnerships and global recognition: Key milestones

Sarvam AI moved quickly after its launch.

 

In December 2023, the company raised $41 million in a Series A round led by Lightspeed, with participation from Peak XV Partners and Khosla Ventures. The funding signalled investor confidence in an India-focused foundational model strategy at a time when most global attention centred on US-based AI firms.

 

In February 2024, Microsoft partnered with Sarvam AI to develop voice-based generative AI tools. The announcement came during Microsoft CEO Satya Nadella’s visit to India, as the technology giant expanded its AI footprint in the country and announced initiatives to provide AI skilling opportunities to millions of Indians.

 

In June 2024, Sarvam AI was named among 10 Indian startups on the World Economic Forum’s Technology Pioneers 2024 list, which recognised 100 global companies applying emerging technologies to address complex challenges across sectors including clean energy, healthcare and biotechnology.

 

By September 2024, Sarvam AI had joined the AI Alliance, a global consortium led by companies such as Meta and IBM to promote open, safe and responsible artificial intelligence practices. The Alliance added seven Indian members, including Infosys and AI4Bharat, reflecting growing participation from Indian institutions in global AI governance conversations.

 

In March 2025, Sarvam AI partnered with UIDAI to deploy an AI-driven solution enabling voice-based interactions for Aadhaar users. The system aimed to collect near real-time feedback on enrolment and update processes and detect overcharging by service providers, with a focus on transparency and accountability.

 

By February 2026, the company reported strong results on benchmark tests focused on document understanding and Indic languages. It said its models outperformed widely used systems such as ChatGPT and Gemini on specific evaluations covering optical character recognition (OCR), document layout understanding and Indic language processing — areas where non-Latin scripts and complex page structures often posed accuracy challenges.

A summit moment, and a broader product push

The India AI Impact Summit 2026 became a public stage for Sarvam AI’s expanding portfolio.

 

Beyond the Sarvam Kaze glasses, the company unveiled 11 AI platforms and solutions in the two weeks leading up to the summit. According to media reports, the company described these offerings as an inflection point for AI applications rooted in Indian contexts.

 

Among them was Sarvam Akshar, built to digitise complex, real-world documents with high accuracy. Sarvam Studio aimed to help creators make content multilingual and accessible across India. Saaras V3, a speech recognition model, was also introduced alongside other AI systems.

 

India-focused models such as Sarvam Vision, Bulbul V3 and Sarvam Arya were designed to handle local languages, cultural nuances and varied user contexts.

 

Speaking at the summit, Pratyush Kumar emphasised the need for India to move “beyond English” in AI. He argued that user experiences differed widely across regions and cultures and that benchmarks must capture subjective and contextual realities, such as voice, vocabulary and lived experience. He said access should be broad rather than restricted by excessive gating.

From vision to scale

In under three years, Sarvam AI moved from incorporation to capital raise, global partnerships, public digital infrastructure deployment and a presence on a national stage.

 

Sarvam AI’s growth came alongside India’s broader efforts to develop domestic AI models and deploy them across public digital services.

 

Whether Sarvam AI would sustain its pace remained to be seen. But at Bharat Mandapam, as policymakers, technologists and entrepreneurs debated the future of artificial intelligence, the startup’s journey from research labs to summit spotlight stood as a marker of how quickly India’s AI ambitions were being translated into products.