With Its Indic Language Arsenal, Can TWO AI Become The GPT Of The East?

There is no better time to talk about India’s quest to create its large language models (LLMs), especially when the world is being dominated by names like GPT-4, Llama, Gork, Gemini, et al. Well, let’s also not forget DeepSeek. Developed with a lean budget of less than $6 Mn, the AI model has ruffled the feathers of many tech giants worldwide.

Now, as far as India is concerned, we are at a crossroads. While VCs and other investors can be seen distancing themselves from foundational models to focus more on building atop the existing ones, the government is advocating the development, control, and deployment of AI using its own infrastructure, data, workforce, and business networks.    

Given the capital-intensive nature of foundational models, which also demand extensive R&D and long gestation periods, the investor concern is not unfounded and neither is the government’s vision to develop Sovereign foundational AI models to compete on a global platform. Hence, this calls for the right balance to be struck. 

Recognising that it was time for India to have its very own foundational models, including interface and reasoning models, when GenAI had just started to create a stir, computer scientist Pranav Mistry founded TWO AI in 2022. 

Headquartered in the US, the startup currently locks horns with top LLM builders of the world in both customer-facing and enterprise AI use cases. 

With its two subsidiaries (in India and South Korea), TWO AI has built models with multilingual capabilities for 50 global languages, which includes all the official Indian languages, along with foreign languages such as Japanese, Korean, Turkish, and many more.

“Our vision has been to enable communication with AI and machines or technology,” Mistry, the founder and CEO of TWO AI, said, adding that the core of TWO AI stems from the idea of unifying humans and machines.

He added that a key driver of this vision is the fact that 80% of the world speaks native languages other than English, and Mistry’s dream has been to make AI accessible to everyone.

In 2023, TWO AI raised $20 Mn in strategic seed funding from Jio Platforms and South Korean internet giant Naver. It serves customers across segments like education, banking and finance, retail, and defence. TWO AI has around 20 paying customers globally, with seven in India.

Interestingly enough, Jio Platforms is TWO AI’s investor and one of the biggest customers in India. With a team of 40 employees, comprising engineers, designers, and innovators from Oxford, MIT, IIT Bombay, Google, Samsung, Microsoft, and NASA, TWO AI is confident of ending FY25 as an operationally profitable company.

The Birth Of TWO AI

Armed with a PhD from the Massachusetts Institute of Technology (MIT) and almost a decade-long stint at Samsung Electronics, Mistry anticipated an upheaval in the realm of AI even before ChatGPT-like models began to dominate headlines across the globe.

Around 2020, Mistry was involved in building AI centres for the company (Samsung) in countries like the US, Korea, Japan, and others. During this time, he realised that a major change was underway, and soon AI was to redefine the world.

Taking cues, he set the sail for his AI adventure. Speaking with Inc42, Mistry said, he spent the first two years, following the inception of TWO AI, building the startup’s GPU infrastructure, training clusters and models, and building an AI architecture. This resulted in rigorously trained SUTRA models to solve two large use cases — for enterprises to scale businesses and for consumers to rely on a vernacular language AI assistant platform.

Today, TWO AI’s ChatGPT-like free AI assistant platform ChatSUTRA (launched mid-last year) enables multilingual conversations.

It has already garnered interest from more than 1 Mn users. Besides, amid growing concern around data privacy, ChatSUTRA promises high security as it runs on regional AI servers.

However, the startup’s models are not open source, which is also where its core monetisation strategy lies. The startup expects an annual run rate (ARR) of $20 Mn in FY25.

Inside TWO AI’s Tech Stack

The startup’s tech stack, SUTRA, comprises three main models – SUTRA-V1, SUTRA-R0, and SUTRA-P0.

While SUTRA-V1 is a pure text-based LLM, its R0 is a reasoning model that competes with larger models like DeepSeek-R1, OpenAI-o1, and LLama 3.1, all while demonstrating better indic-language capabilities.

Meanwhile, TWO AI’s SUTRA-P0 is a predictive model, designed for forecasting and analysis. Predictive models have nothing to do with language; they are purely time-series models, Mistry said. 

“In enterprise AI use cases, chat applications make up just 5% of the scope. Real AI optimisation happens in workflows and manufacturing lines, where predictive models play a crucial role.”

On top of its foundational LLMs, TWO AI has built multimedia models such as a voice model called HiFi and an avatar model.

TWO AI’s Way Forward

Mistry told Inc42 that he does not aspire to become the OpenAI or Perplexity but rather Palantir Technologies of the East, which can enable AI solutions at scale for large enterprises, eventually reaching real customers for real-world applications of AI.

While the US, Korea, and India are its main markets, TWO AI aims to scale its offerings across the APAC region, barring China.  Mistry has a special focus on India. 

This is because, as per Mistry, there are no India-based search engines or messaging platforms. To address this, TWO AI is also working on launching a homegrown consumer application, much like Yahoo! Japan or China’s Baidu. The launch is expected in the next few months. 

Meanwhile, Mistry’s plan for its enterprise AI solution is simple — not to acquire thousands of customers but the world’s top of the rung. The startup aims to increase its global customer base to 25 to 30 clients by next year.

According to Mistry, he does not wish to raise more capital, as the company has already reached the stage of operational profitability. Even cash burns are within limits. 

Now, as GenAI continues to make its way into basic workflows and operations across sectors, India needs to build its own foundational models for security reasons and India-specific requirements.

Though Bhavish Aggarwal’s Ola Krutrim, PeakXV-backed Sarvam, CoRover, and a few others have made strides in building LLMs and small language models focussed on indic languages, there is a need for more endeavour in this area. 

While TWO AI has been able to build foundational models that are more closely competing with the tech giants globally, TWO AI has plans to join hands with the Indian government to build open-source Sovereign AI models.

However, Mistry believes that the noise around AI in India is still disoriented and experts need to focus on solving more India-specific problems with more modular models with a high level of collaboration. 

“No one is wrong in AI. Everyone needs to work together. India needs a lot of solutions, and not one LLM solving it all,” he said.

For now, it remains to be seen how Mistry walks the talk of capturing the broader APAC market while helping India with a more sophisticated indic language-based AI assistant platform that promises unprecedented data privacy and context.

Now, as GenAI presents vast opportunities — with India alone expected to surpass $17 Bn by 2030 — can TWO AI challenge the dominance of giants like OpenAI, Google, Meta, or DeepSeek?

[Edited By Shishir Parasher]