Vishal Sunil, Co-Founder and CTO of Rocket Learning on Using AI to Scale Early Childhood Education Across India

Yeah, I think yeah, exactly right, I think the keyword there is excitement. I think tech has been impacted by the end rate of AI and has been trying to be more reactive rather than proactive right at least that's the feeling right because now every student out there is going and using it as a crazy statistic from the US is that you know 39% of five-year-olds have interacted with some Jenny at YouTube. Hello and welcome to the do-unbetter podcast and philanthropy sustainability and social entrepreneurship. I'm your host Alberto Lieggie from London. Please click that subscribe button and follow us if you're not already doing so and do leave us a rating and a review. It helps us to find the show as well. Today it's an absolute pleasure to welcome onto the show V-Shall Sunil co-founder and CTO of Rocket Learning. Rocket Learning is based in India. They're engaging with over three million stakeholders across 10 different states in India and today we're looking at the work they're doing on early childhood education by embracing artificial intelligence. They have a really interesting app called APU that is helping parents and children alike have better engagement, better interaction, better learning outcomes, better stimulation. So you're in for a really interesting conversation and on that note V-Shall a big heartfelt welcome onto the do-unbetter podcast today. Thank you so much Alberto. Tell me a little bit about the work that you're doing at Rocket Learning. You're one of the co-founders and you're also the CTO, a chief technology officer. What's going on at Rocket Learning since the last time you guys were on the show two years ago? Amazing thanks. I think we remain as committed to community-based learning, right? But we do want to build on the impact and deepen it, right? So that's something that we're very keen on just as we're scaling the impact, right? So I think maybe two years ago we were working with a million students today. We'll be working with close 3.2 million students across 10 states in India and have scaled up this community learning as well as you know teacher development. So the products that we've built to you know many, many state governments and district implementations, right? So I think Rocket Learning, our commission is that every child you know reaches their full potential, right, as they go into schools between the ages of 0 to 6 actually. So we've increased our ambit of even working with younger children and making sure that development we can have them be on track, right from when they're born, right? So I think this is the unique space in India where everything is very nascent, right? You know users are nascent as well, but I think you know the system is the systems are now evolving very quickly to you know emphasize early childhood education and bring about a change in you know the daycare workers being retrained at teachers, right? For example, the cross India is a mission that we have actually pushed at the central government level where we've been able to launch a campaign called push in B but I B which means not only nutrition but education as well. So we've managed to get you know about 1.5 billion dollars allocated to that for retraining these teachers, right? So I think a lot of exciting things happening at Rocket Learning and now we're 500 people but we hope to still fold that startup mentality. It's interesting how you well 500 people that's not a small amount and we're going to we're going to drill into that but you have the privilege in some ways of working in India where the sheer size of each individual state makes a lot of our countries here in Western Europe, for instance, we're based. You know seem rather small and I remember you guys talking about you know running a pilot at scale but having this entrepreneurial mindset having a pilot but piloting with like a million people. So you have 3.2 million students that you're engaging with across 10 different states in India and okay 500 folks within the Rocket Learning family. Before we drill into some of the exciting stuff that you guys are doing give us a little bit of a flavor of what it is that you guys do. So we got the focus area tell me a little bit about how you're approaching what you do and then we can drill into some of the details. Absolutely so the idea is somewhat to have continuous support for two things right now home based learning involving the parents in the learning of their children right in these constraint systems and to upskill these day care workers as preschool teachers in India has the worst largest day care system. It's called the Allen Valley system which has about 1.4 million centers across India right and so basically we want to super charge this force and about 60% of India's students go to these day care. So if we can bring about some enrichment and learning there I think before we intervene only 15 minutes of instruction were happening at these centers and I think our mission has been to move that along and equip these teachers through continuous training right. So I think the philosophy of both our products is one easy to access low low low tech in terms of their understanding in terms of their need of you know downloading a new app or a new interface right so we still are committed to what's that we do community based learning right so as opposed to one on one learning to a chatbot or an app that you know people tend to all focus on personalization in that case. We focus on engagement in a very you know radical way which is that if you have a community of those neighbors around you the fact that one person doesn't assignment and sends it back what awaits the whole community that they could do it right. So the community based education model is common both in our teacher and our parent model so through this sort of you know holistic approach where we're doing policy upscaling the teacher and the child we're hoping to drive impacts and and one of the sneak peeks is that I think soon going to launch some results of our CT were showing that you know the up learning intervention is almost you know delivering the same amount of standard deviation. In fact as you know a whole new young and what you work a whole new day care worker in the system right so that's some a little bit about great doing it and I'm looking forward to reading about that randomized control trial that'll be pretty cool and for the listeners as well I think there is a case study at Harvard Business School on rocket learning which is also interesting and just to touch on on something you mentioned a second ago which is about what's up. I guess that was sort of part of the conversation from two years ago by not reinventing the wheel and need and yet another social media app and saying look what's up in itself is good enough for what we need and for the some of the user generated stuff which is which is cool and now over the last two years times have changed radically when it comes to artificial intelligence AI. What's your you're the CTO what's the world looking like tell us a little bit about the products and what it is that is keeping you up at night with excitement. Yeah I think you're exactly right I think the keyword there is excitement I think attack for a lot of for for this period has been impacted by the end rate of AI and has been trying to be more reactive rather than true active right at least that's feeling right because now every student out there is is going in using it as a crazy statistic from the U.S. is that you know 39% of five year olds have interacted with some generic right from the common sense media report. Right so I think I think the students are adopting this really quickly however systems are still sort of you know shy and dealing with this and being proactive about this. With only eight percent of teachers in the U.S. you know feeling that Gen. A.I. is productive in in that last right so the huge divide that right so I think that's that's where the industry is and I think rocket learning what we wanted to do is take that first proactive step to show that look generative AI can be a really really powerful source of providing high quality personal education. To two children you know in this age group right because children between the ages of three to six naturally they don't like following linear flows they don't like the quizzing they don't like you know they they love to chat with you they love to talk they love to discover through. Seeing the world around them and objects around them there's a real joy to the learning right so I think a lot of the attack try to bottle them into this you know control environment right and then say that okay yes you have to you know go from A to B only through this you know these few clicks of the button. of answering these few questions. So what we wanted to do is reimagine that. So in the last six months, along with Google.org, they send us about eight Google engineers through the Google Fellowship Program. And our team and their team together came out and through designing and research have created a pedagogy first voice tutor called Appu. Where, you know, it's just a link that you can play. It's a web app. It could be an app someday. But the idea is that you have a two to three minute conversation with this fun elephant, right? And that fun elephant teaches you one concept in that session through everyday objects around them or through call and repeat, rhyme, stories, and so many, so many different non-linear sessions. Excellent. And that's a vote of confidence if you're getting the likes of Google.org to deploy some of their engineers to give you a hand, which is pretty cool. And so somebody listening to this might be thinking, OK, but what does this look like? Do you grab a tablet and stick it in front of the child? Does the teacher or the parent use the AI as a sort of guide to help them interact with the child in a more, I don't know, more knowledgeable? It's a great question. Yeah, that's a great question. So I think these are the things that we started off with. Actually, we worked back from a user problem, right? And that's always something I'd re-emphasize. The user problem was that I think, yes, there's most ethics. After six to nine months, what we find is that users tend to saturate the learning, right? And their experiences and tend to drop off. So if you continue with just a group-based model, without any personalization, what we find is that most parents are like, look, my cell is so sharp, they've already finished all of your content, right? So I think, especially the super users, right? So I think how do you reinvigorate them and make them go the next level, right? That was a key problem that we wanted to address. Second thing is that I think we tried like games, tap games and those sorts of things. What we found is that they don't, I think, while you can drive some amount of engagement and find out of them, we didn't feel there was great learning density there. And we also didn't want this to be something that had high screen time, because philosophically, we wanted to stay away from that. So why we chose a voice-based thing, where your need to be focused on the screen is almost secondary, right? But rather, it is a conversation that you have with this elephant, right? So what we build is sort of that in one particular-- we've sort of got drilled it really well and designed the learning session in a way that, you know, you could have one particular link, so one particular learning module out there. And Appu, our Genai, our elephant tutor, will only stay on that topic, right? Appu will have many examples from our content back on how to teach this particular topic. Let's say it's content, right? So Appu would have access to a pre-verified asset bank of, you know, images, a pre-verified asset bank of objects that are common in India, right? And use both the reference of these instructions on how to teach the instructions on examples of, you know, the modus of teaching, accounting, you know, call and repeat those sort of things, and the asset bank to put together a unique learning session that would last two minutes, right? So it could just start off by asking you, hi, how are you feeling today? Are you excited to learn about numbers? And the student could say no, and Appu could, you know, try and, and say-- and try and get them excited about it, right, once or twice. And if this child continues to say no, we end the session there, right? But if we do go forward, you know, we could even ask them, you know, what is your favorite fruit around there? And then try to teach them with, like, you know, the favorite objects around them, and make it more personable to them, right? So that way, numeracy accounting, all of that happens dynamically, and you really get a one-of-a-kind experience. So you try to make a really bad real experience. There's a two-minute conversation that delivers a really enriched, joyful learning, and all of this comes from, comes back to the point about human-centric AI. So how we designed it is human-centric, yeah, that's right. - Fascinating. - I know we've gone a little long. There's one small other point that I wanted to make, which is that, you know, what we, one of the things that we noticed, right, and this is something that we realized after we went on field, and so at Rocket Learning, we go on field with our products every month, right? So every month of development we go on field, we test it out. There's no point building in a lab and figuring something out later, but what we found is that a couple of things, is that like the parents were, so we always recommended the parents sit with the children, right? While these sessions are going on, right? And they helped the child, right? So we don't like the idea of giving the child the phone. So the recommended is that the parents sit with the child and has a conversation together. Now what we found is that the way our voice detection models were working, it was detecting the parents voice. Parents were trying to teach the children while this lesson was going on, right? And it was often marking them as wrong, right? Like, oh, it's saying, no, no, why don't you just, I mean, yeah, you're right, but you're right, asking me another question, but why don't you give me the answer to this or things like that, right? Which, which I felt that, okay, yes, this is something that is also a place where we have to change a tech instead of a place where we have to change the parent behavior. Because this is really, really what you want to foster. You want to foster better conversations between mother and child or parents and children, right? And have those natural learning moments at home. So it's up to us to then change the detection and split out the child's voice and the parents' voice. And so up to just be a part of the conversation and not necessarily be there to, you know, say you do this right or wrong, right? So it's something that was interesting, example of human-centric design. - Yeah, that's fascinating. How you're, you're ensuring that the parent is always fully engaged, which is, which is great. Now, India, there's so many different languages. So you said you're across 10 states, 3.2 million. What language is Abu speaking? Does Abu multi-lingual, does Abu speak English? And then if Abu does speak English, could somebody use him or her, I'm not sure Abu's gender, but even if they're not based in India? - Yeah, it's a great question. Actually, you mentioned too, right? That each state is almost like a country in the European Union and India. And that's exactly the point, which is that there's so many different cultural context and languages across these states. We currently work in four languages, right? So we've started off our fourth language. Actually, fifth language this year, right? She comes English as well. But yeah, I think we have many implementations in these languages. The brilliance of JNAI meant that by adding one sentence to the prompt, we were able to deliver it in English or Hindi or Marathi or any other language. So that's the great thing about these models, right? Yes, they do require guard railing, tuning, all of those things. But I think the great thing about these models is fundamentally you can instruct them in English. But they have a good amount of knowledge bank in all of these languages, at least the big languages, right? So between, I think, 20 to 30 languages across the world are quite readily supportable by doing this. Now, the one thing that we do do afterwards is that we do want to see whether the vocabulary used by the bot in that particular language matches the vocabulary that is generally used in the classroom, right? So there is that little bit of fine tuning that we do expect to do for each language. So we do tell it. So for example, like the idea is to call it activities and play and those sort of things in our childhood, right? In English and that should reflect in Hindi as well. It should reflect in another language. Yes, I've proved it's available in English. And I think we're hoping through the next few months to launch a publicly available version that sort of we are confident and we can stand behind and hopefully everybody can build on top of it somewhere. Right, so it's not a question of going to the app store just yet for most listeners, but conceivably, a little bit farther down the line, you could be in Brazil and maybe have a Portuguese speaking up. That's not inconceivable. - No, it's not at all, but I think generally how we've designed it is to be very, very highly in sync with the content that we are teaching, right? In our product. So it was very much build it for a purpose rather than build it as maybe somebody will adopt it. What we do wanna understand at the end of this academic year is that we hope to prove that it has an amazing learning outcome impact as well, right? So that's spending six months with the AI tutor a couple of sessions a week, lead to learning games, right? What we're seeing in early studies across, I think in a few places in Africa and Nigeria and so on, right? Through these chat CPT studies is truly amazing, right? Just giving access to some tools like this, right? But I think for the early years, it's still a question mark and doing it through these got real learning sessions is also a question mark, right? I think so I think both of these we want to prove out that it does work and I think at that point, we really get on to the, you know, our, you know, wish of making this as open source and adoptable across the world as well. - That's quite a few foundations who are focused in early years and education across the world who I know and who I think would probably be interested in exploring that. The randomised control trial that she mentioned is coming down the pipeline, when are you expecting that? - I think, you know, more than a question for us, it is a question for our research partners, but I think, you know, these things take time because they do go through a lot of reviews and feedback and verification. So, but our optimistic hope is that, you know, in July, we're coming out with this topic there. - Okay, okay, along with J-PAL. - Yeah, I'd love to, well, we have the folks from J-PAL and from Betis Foundation as well, who I think are supporters. - Yes, are supporters, yeah, absolutely. - So shout out to Murugan on that front. - Yes, so. - The journey for you, how did you end up where you are today? CTO, co-founder, dealing with millions of stakeholders across such a huge geographic area and doing something that's so consequential, not just for SDG for an education, but on broader areas as well. - Yeah, absolutely. I think, you know, for me, it was, you know, very quickly, I was a data scientist, you know, right when the term was invented in the early 2010s in the US, and very soon, I figured out that, you know, entrepreneurship is something that I'm very excited about. So, I have a couple of firms there in Boston, particularly one that I was working on was, you know, AckTech trying to detect pests in, you know, pepper farms in Florida through drone imaging, so building some deep learning models there to try to, like, model best damage in plants. And, yeah, I was sort of starting for a purpose for, you know, my particular skills of, you know, liking mathematics and computer science from, since I was like 13 years old. So, I think, you know, I found a particular program called Development Economics and Data for Power, Power Day alleviation, that was sort of a first MIT X program that J-PAR launched on sort of how we could use, you know, evidence in policy-making and how you could use some of the, you know, latest statistical method, I think, you know, asset of flow, and professed of flow, and that's a vanity, that's a vanity, also, you know, ended up building Noble Pite a couple of years after I finished that program, but for their work, so I think that sort of taught me very excited about contributing. And for me, what drives me every day is that I love building products for people, for underserved people, right? So for me, that's what excites me on a daily basis that I wake up and I say, okay, look at this mother, look at this mother who, you know, the family has one smartphone, right? The smartphone is taken away by the father who works, comes back 8 p.m. this mother waits every day by the door at 8 p.m. takes that smartphone, goes to her of learning WhatsApp group, does these videos with the children, and then sends it over that night, so that her child can get all of the smileys and recognition on the group, right? So I think, and ultimately through that, you know, become the best person they can be, right? So I think stories like that, and having an impact like that through your product really has kept me here for, I think, now the five years that we've been, you know, doing our work at Rocket Learning. - Excellent, excellent. Well, as I see, you've found your calling, which is absolutely great, and I mean, there's so many different things that are going on on the philanthropy and social impact space in India, that I think also you're not only fortunate in terms of having that space in which to pilot and test things that, but also I think such a rich ecosystem of other innovative thinkers, which can only vote well for you. Before you run off, I'd love to ask you for a key takeaway. What's that one thing that you'd love for the audience to keep in mind after they finish listening to today's episode? - Yeah, I think, you know, I just thought we should be more proactive about this era of AI, and believe that, you know, human-centric AI can be a huge, you know, hugely positive force in education. You know, 17% of all chat GPDs, where these are related to education, and education is the number two sector, right? In terms of adoption of gendered AI. So I think there's a huge opportunity for us to, you know, deliver outside the impact that scale through gendered AI, and I want to leave us with that positive message, that, you know, it's possible to, you know, create a highly, highly impactful, deep learning, God, real experience for children. - Wonderful, and on that positive note, Vishal, thank you so much for joining me and joining us on the Do One Better podcast, and for bringing Rocket Learning back onto the show as well. It's been two years since you guys were last on the show, and I'm looking forward to hosting you back at a little bit down the line as well. So thanks so much for making the time and for all the insight, very, very enjoyable and very insightful. - Thank you so much. - Perfect, and that's a wrap. Thanks very much for tuning in. As always, you've been listening to a great chat with Vishal Sunil, co-founder and CTO of Rocket Learning. For information about this conversation, and more than 300 other interviews and case studies with remarkable leaders in philanthropy, sustainability, and social entrepreneurship, just visit our website at leegee.org. That's l-i-dj-i.org. Please click that subscribe button and follow us if you're not already doing so, and do leave us a rating and a review. It helps others to find this show as well. Thirdly, enjoy producing today's episode for you, and I very much look forward to catching up with you again this coming Monday. (upbeat music)