India should stop playing catch-up with the West on building large language models and instead look ahead to what’s next, says Pranav Mistry, founder TWO AI, the maker of SUTRA AI models. TWO AI counts Reliance Jio as one of its investors.
“The next wave of models will be quantitative models and world models. These will work with structured data like transactions, weather, traffic, soil, and visual data. That’s where real AI impact will happen, not just chatbots,” he said.
Quantitative models, or LQMs, will power applications in governance, agriculture, public infrastructure, and cybersecurity. “These models can predict patterns from data even in the wild, data that hasn’t been seen before. It’s not just about training on past corpora. It’s about making real-world predictions from real-time data,” said the computer scientist who owns 100+ patents in the field of AI, robotics and AR/VR.
His visionary academic project at MIT called ‘Sixth Sense’ became pivotal in shaping Samsung Galaxy Watch, foldable display tech and virtual reality while Mistry was leading Samsung Technology & Advanced Research (STAR) Labs.
He was cautious but optimistic about the Indian government’s AI mission, which includes compute subsidies and grants for foundational models. “The heart is in the right place. But we must stop copying the US or China. Six months later, by the time we build an LLM, the world will have moved on. We need to think ahead, not follow,” he said. “The government should focus on where the world is going, not where it was.”
TWO AI is considering submitting a proposal to build a fully India-trained version of Sutra under the government’s open call. But Mistry cautioned against simply throwing money at model training. “India’s AI future lies not in mimicking global models, but in building an India-first technology stack that is designed with local needs, languages, and scale in mind.”
Ultimately, Mistry believes India has the talent and potential to lead in AI but needs strategic leadership. “Just saying ‘India will be number one in AI’ doesn’t make it true. We have to build it thoughtfully, securely, and for our own people,” he said.
Mistry urged that all model companies who are aiming to expand in new markets like India must commit to data localisation. “All our enterprise models run on servers in Mumbai. That’s important. Enterprises, especially in banking and government, don’t want their data going outside India. And they shouldn’t,” he said.
TWO had previously raised $20 million from Reliance Jio and Korea’s internet giant Naver Corp. It is looking to raise another $35 million - $50 million round. The company has an annual recurring revenue of $10 million and is expecting to turn profitable by 2026. “Our vision aligns with Jio to build AI for India. Jio understands that global models are not India-first. We work closely with their leadership, not just for business, but to shape the future of AI in India,” he said.
He said that TWO does not aim to compete with OpenAIs of the West or Krutrim, Sarvam of the East. “We don’t want to become a model company. From the very beginning, TWO never intended to compete with OpenAI or Mistral. Our business is not selling models for $1 per thousand tokens. Our core offering is solutions that enterprises can deploy securely and at scale,” Mistry said in an interview.
The firm is set to launch the Sutra AI Studio for enterprises to build AI apps and agents. Mistry revealed that TWO will soon launch another consumer-facing product called Geniya, a free-to-use India-focused search engine and AI assistant akin to Perplexity AI.
“Geniya is already live in beta. It’s not officially launched, but people are using it. We plan to go public with it this summer,” he said. “It’s built for India’s search and knowledge needs. It will let users ask questions in any language, even through voice.”
TWO AI’s new reasoning model, R0 has garnered 1 million users within 30 days and over 2,000 developer signups. He added that Enterprise adoption is growing rapidly. “We have recently signed two large enterprise clients in Korea. And in India, we’ve received greenlights for government projects related to mission-critical applications.”
“The next wave of models will be quantitative models and world models. These will work with structured data like transactions, weather, traffic, soil, and visual data. That’s where real AI impact will happen, not just chatbots,” he said.
Quantitative models, or LQMs, will power applications in governance, agriculture, public infrastructure, and cybersecurity. “These models can predict patterns from data even in the wild, data that hasn’t been seen before. It’s not just about training on past corpora. It’s about making real-world predictions from real-time data,” said the computer scientist who owns 100+ patents in the field of AI, robotics and AR/VR.
His visionary academic project at MIT called ‘Sixth Sense’ became pivotal in shaping Samsung Galaxy Watch, foldable display tech and virtual reality while Mistry was leading Samsung Technology & Advanced Research (STAR) Labs.
He was cautious but optimistic about the Indian government’s AI mission, which includes compute subsidies and grants for foundational models. “The heart is in the right place. But we must stop copying the US or China. Six months later, by the time we build an LLM, the world will have moved on. We need to think ahead, not follow,” he said. “The government should focus on where the world is going, not where it was.”
TWO AI is considering submitting a proposal to build a fully India-trained version of Sutra under the government’s open call. But Mistry cautioned against simply throwing money at model training. “India’s AI future lies not in mimicking global models, but in building an India-first technology stack that is designed with local needs, languages, and scale in mind.”
Ultimately, Mistry believes India has the talent and potential to lead in AI but needs strategic leadership. “Just saying ‘India will be number one in AI’ doesn’t make it true. We have to build it thoughtfully, securely, and for our own people,” he said.
Mistry urged that all model companies who are aiming to expand in new markets like India must commit to data localisation. “All our enterprise models run on servers in Mumbai. That’s important. Enterprises, especially in banking and government, don’t want their data going outside India. And they shouldn’t,” he said.
TWO had previously raised $20 million from Reliance Jio and Korea’s internet giant Naver Corp. It is looking to raise another $35 million - $50 million round. The company has an annual recurring revenue of $10 million and is expecting to turn profitable by 2026. “Our vision aligns with Jio to build AI for India. Jio understands that global models are not India-first. We work closely with their leadership, not just for business, but to shape the future of AI in India,” he said.
He said that TWO does not aim to compete with OpenAIs of the West or Krutrim, Sarvam of the East. “We don’t want to become a model company. From the very beginning, TWO never intended to compete with OpenAI or Mistral. Our business is not selling models for $1 per thousand tokens. Our core offering is solutions that enterprises can deploy securely and at scale,” Mistry said in an interview.
The firm is set to launch the Sutra AI Studio for enterprises to build AI apps and agents. Mistry revealed that TWO will soon launch another consumer-facing product called Geniya, a free-to-use India-focused search engine and AI assistant akin to Perplexity AI.
“Geniya is already live in beta. It’s not officially launched, but people are using it. We plan to go public with it this summer,” he said. “It’s built for India’s search and knowledge needs. It will let users ask questions in any language, even through voice.”
TWO AI’s new reasoning model, R0 has garnered 1 million users within 30 days and over 2,000 developer signups. He added that Enterprise adoption is growing rapidly. “We have recently signed two large enterprise clients in Korea. And in India, we’ve received greenlights for government projects related to mission-critical applications.”
