The transformer birthed GenAI. Meet the man who built it

Mint

2 days ago

Eight years ago, Ashish Vaswani led a team of Google Brain researchers that invented the transformer, the magic sauce behind generative AI. The new model, which learns and generates human-like text, took the world by storm as it made its way into ChatGPT, and later, others like Gemini, Grok and DeepSeek.
India-born Vaswani, whose startup Essential AI works on building foundational models, now wants to establish an India team, access graphic processing units, find clients and spot a local strategic investment partner.
'We're building foundational AI models to automate coding and tasks in science, technology, engineering and mathematics (STEM) applications. The core idea is to build the best models in specific fields, so that we can then partner with large clients who will licence our models to build applications on," Vaswani said in an interview.
'Could have never imagined it'
In June 2017, Vaswani led a Stanford University research funded under Google Brain, floating the transformer model now regarded as one of the world's most significant inventions in computer science—alongside the likes of Frank Rosenblatt's neural networks, and Sergey Brin-Larry Page's PageRank.
Asked whether he expected the model to have the kind of impact it did, Vaswani, who is in India after 16 years, said that he 'could have never imagined it."
'What I had set out to build was a better version of machine learning, and improving the way machine understanding worked. I never thought it would explode into what it is today, and the way it has taken over our lives."
Also read | Generative AI, data centres to define India's tech industries in 2025
Vaswani, 39, is chief executive of Essential AI, which he co-founded in 2023 alongside Niki Parmar, a co-inventor of the transformer model. He stayed on at Google until 2021, leaving to build Adept AI Labs—a platform that today has a licensing deal with Amazon to build its AI initiatives. Vaswani and Parmar left Adept in less than two years over reported differences with investors, and started Essential AI.
Fundraise
GenAI burst into prominence when Sam Altman-led OpenAI, a Silicon Valley peer of Vaswani's, unveiled ChatGPT in November 2022. Since then, AI has become a household term, catapulting the field into prominence well beyond engineers and researchers.
His startup, which raised $56.5 million in December 2023 and counts AMD, Google and Nvidia as investors, will be looking to raise a second, larger funding round of around $100 million later this year, Vaswani said. 'The results of our early foundational models are here, and they look good. We'll be using these results as a reference point for our next fundraise," he said.
As part of this move, Vaswani is open to interest from Indian strategic partners as well. 'India has some of the brightest minds, and it is absolutely important that India pursues building its own AI. There's no reason why foundational work in AI cannot happen in India," he said.
Read this | Mint Explainer: What OpenAI o1 'reasoning' model means for the future of generative AI
Investors negotiating with global ventures concur, stating that foundational work in AI will have the scope to differentiate the work on GenAI that ventures across India and abroad are pursuing.
Foundational AI
'One has to look at a big enough problem, and assess how many millions of people a problem impacts," said Anand Daniel, partner at venture capital firm Accel. "Then, we look at the solution being built, and the foundational engineering that a venture is undertaking in order to build for the problem. It's still early days, but the scope for foundational work remains broader in the US, than what Indian startups have so far created," Daniel added.
Both agree that there is room for ventures to exist even in the foundational engineering space in GenAI in the long run, despite a battle for dominance playing out in the US among the likes of Google, Microsoft and OpenAI.
'I fully think that there is enough space now to build products and companies that exist alongside and outside the Big Tech environment, and that will further widen as generative AI evolves. Eventually, there is ample scope for many to disrupt the global technology environment," Vaswani said.
And this | India's generative AI startups look beyond building ChatGPT-like models
Foundational AI, to be sure, is seen as a tough to crack since it requires firms to build and train their own algorithms from scratch. While the advantages include the ability to have a proprietary AI model that squarely targets a specific use case, doing so requires significant working capital, a key challenge in India.
Capability questions
Vaswani, felicitated as one of India's 30 leading minds in AI by Accel in Bengaluru on Wednesday, is based in San Francisco. While Essential AI is headed by Vaswani and Parmar, the core team is in the US, highlighting the country's lack of focus on core engineering driven by access to capital being much lower than in the US.
'This is certainly an issue, and core engineering capability continues to lag in India. This is one key factor that we're also looking for in startups, but a lot of work happening here goes amiss in terms of core foundational work. Strategic companies doing foundational work will be key to progress in the field," added Prayank Swaroop, AI investor and partner in Accel.
Vaswani, however, said the evolution of GenAI likely has to do a lot with philosophy, alongside computer science and mathematics.
'Is computer science more mathematics or philosophy? It is perhaps both. Steve Jobs was the first person to articulate that successful products are a blend of technology, the liberal arts and philosophy. This is what can lead to us doing visionary work. Eventually, we're building the philosophy of how the world should be. The ethos behind technology is to solve problems, and that's the only job of innovation," he said.
And read | Gen AI pushes global firms to pour money into hardware upgrades