‘Seeing many light wrappers pose as AI solutions’: Google India startup accelerator lead

While the programme for the apps-only cohort of 20 startups runs from July to November, the AI-first cohort participate from September to December.

“The COVID-19 pandemic saw a trough because the number of people starting up and the number of companies out there was kind of plateauing out, but since the AI shift, we have seen a huge spike in the number of startups who apply to Google’s accelerator programmes,” Paul Ravindranath G, senior programme manager, Google for Startups Accelerator India, said.

“The reasons are, people are able to utilise AI tools and AI models to enable their solutions in a much quicker and a faster way,” he added. In an exclusive interview with The Indian Express, Ravindranath further reflected on Google for Startups’ journey so far, the state of India’s startup ecosystem, the most common challenges founders encounter when building AI solutions in the country, and more:

How has Google helped shape India’s startup ecosystem, and what impact do you think it has had on startup founders across different sectors?

Ravindranath: I’ve had the privilege of being associated with the startup programme from day one for Google in India. At the beginning, it was just a one week-long mentorship programme. We would invite about 10 companies together in the office and interact with them about their needs and multiple areas such as technology, growth, marketing, etc.

2015 was more of the era of “Me Too” kind of startups. I remember Google was the foundational partner for the 10,000 Startups initiative by NASSCOM. What started as supporting founders to grow and innovate has snowballed over the last decade into so many meaningful areas of value. Over the years, we have seen India’s startup ecosystem evolve not just in terms of volume of ideas, but in solving massive population-scale challenges on the back of platforms like Google Cloud and other AI tools we have out there.

Today, we stand at over 237 companies supported across 17+ cohorts. Over the last four to five years, we’ve seen startups building deeply on AI and leveraging technology in incredible ways. These companies have gone on to be really successful as well, raising more than $4.5 billion in funding, with a 96 per cent survival rate. It’s a real testament to the fact that these companies are solving real problems and doing it in a way that allows them to sustain and grow, with support from platforms like Google that are able to back them.

How does Google decide which startups make it into its accelerator programme? How has the selection criteria evolved over the years?

Ravindranath: In the early days, we were looking at startups in what was really the era of “Me Too” companies. We had lots of applications, several at the idea stage, wanting support from our programme. During the COVID-19 pandemic, over those two to three years, we didn’t switch off this programme. It remained very much active, we pivoted it completely online.

That was a challenging period for the Indian startup ecosystem. The needs of companies and the way we selected them also changed. In our cohort at the time, we picked companies focused on healthcare or access solutions, and we saw how tech enabled those solutions.

The pandemic was also an interesting phase where startups in India saw a complete shift in their user base. My dad, who is over 70, had never used an app other than YouTube on his phone, and by the end of the pandemic, he was ordering groceries online, consuming online content, and fully engaging with the e-commerce ecosystem.

In the last four years, with the leaps we have made in AI capabilities, we’ve seen a complete platform shift. The number of startups using AI to solve population-scale problems has grown rapidly. We are seeing strong trends in healthcare and fintech, where AI platforms are being used to address these challenges.

This, however, presents another challenge for us. How do we choose which companies are truly building with AI? We see many “light wrappers,” as we call them, startups or aspiring founders building quick UIs on top of existing large models.

That’s not something we want to support, because those are easy to spin up and not necessarily sustainable solutions. We go a level deeper to see how startups are tackling interesting problem areas using proprietary data, applying AI and ML to that data to personalise and improve accuracy. That shows promise and gives us a chance to work with them to make their solutions even better.

How are you seeing Indian startups use AI tools to build solutions for India?

Ravindranath: I’d love to maybe recount two or three that have really stuck with me over the years as being truly impactful. One of them is a startup called Niramai, which is Bengaluru-based. Gita Majunath is the founder. They are in the early breast cancer detection space. So imagine with a thermal image of a woman’s chest, and applying their proprietary AI model with a 98-plus percent accuracy, they’re able to detect early-stage breast cancer.

Another startup that I really love working with is this startup called Wysa. Wysa is a startup that actually operates in the mental health space. It began as a simple AI chatbot that you could talk with if you’re feeling down and low. But the AI chatbot has evolved, and clinically cleared to be able to get you out of a tough spot.

Kidzovo is an app that we worked with in our cohort last year. It is a great example of a startup that adds a layer of interactivity over a video. So let’s say a young child is really spending a lot of screen time, consuming a lot of YouTube videos, cartoons, Peppa Pig, and the like. With AI, Kidzovo is able to add a layer of interactivity. It will pause the video the child is watching and ask them an interactive question: “Can you tap on the nose of the pig? Can you tell me the number of sharks in this Baby Shark video? Tap on all the sharks.”

Another one I can cite is a company called Toon Sutra. They are using Google’s Gemini models, Veo and other video-generation models in very interesting ways to create India-focused comics.

As AI becomes more central to startups, what are the biggest challenges you see for early-stage founders in India?

Ravindranath: One of the biggest challenges is compute. A lot of AI founders working in the space look for support in subsidising the cost involved in building AI solutions. Often AI solutions involve a lot of data processing, and as founders, they definitely try to strike the balance between a quick AI solution that might be very costly versus something that is cost-effective and delivers the right result.

Doing this cost-benefit analysis as founders and CTOs is really important. This is one of the areas where, as an accelerator, we help founders architect for scale. We have very specific modules in our programme that evaluate the package architecture, evaluate the models and their accuracy, and determine how expensive a solution could be, and how to make it more efficient.

Another challenge I’d like to highlight is how to go through this current AI boom and do it responsibly. Data is available everywhere, and people can quickly build solutions around it. But how do you build thoughtfully?

If you’re building in the healthcare space, in finance, or in tech, as founders you need to develop a mindset of responsible innovation. How do you treat the data source responsibly? How do you work with that data and information? How do you manage the interaction between people and AI? I think that is really important for founders to focus on.

With the rise of vibe-coding, how do you think this changes the way Indian startups are building, experimenting, and scaling new products?

Ravindranath: There’s now a huge increase in the ability for entrepreneurial minds to prototype solutions. So anybody today who has a great idea, in the past, they would have to figure out how to go about solving for that: figure out getting somebody else, like a co-founder, to come on board, think through it, model the solution, do UX research, go on the ground, talk to potential users, create a minimum viable product, work on initial traction, get user feedback, and go through multiple cycles involved in that innovation.

But today, with platforms like Firebase Studio or other vibe-coding platforms, we are able to get to that initial prototype or desirable MVP in a very, very short period of time.

So the innovation cycles are much shorter in today’s day and age. Now, that’s good because with a lot of the ideas, we are able to fail fast and move forward. But at the same time, it creates a problem where a lot of support programmes out there are trying to design filters, which is a thin wrapper, which is not a real AI solution. As an ecosystem, we’re all learning to find those gems, but overall, it’s a positive movement towards innovation.

At the same time, I think it needs to be approached with caution, because as founders are building, one of the questions that I get is: “Hey, where do I find some really great AI and ML or big tech engineers to help me build out this product?”

So while vibe-coding is great for getting to that initial vision and prototype, to build a truly sustainable, production-ready solution requires investment in foundational knowledge, having a team with deep tech expertise to really build that idea out into something production-ready that can scale at population scale, especially if you are solving for India.

Any AI startups supported by Google that are focusing on building foundational AI models or native AI apps?

Ravindranath: One company, called Deep Vision Tech, a Bangalore-based company, is working in the live translation, sign language space. They have an avatar on their phone and, as I have a conversation with someone who cannot hear, the avatar actually signs to them in real time translating and signing back to them in Indian Sign Language. And there are multiple sign languages. This is an amazing innovation in the accessibility space.

Another startup is called Iyaso. This startup is in the speech therapy space. The founder himself has a stammering problem, and he developed this app to actually help people who stammer overcome that disability, with a great degree of success and clinical validation as well. He technically has one of the largest databases of stammering data and is approaching the problem entirely with an AI mindset and methodology.