# Students Will No Longer Be Taught By Teachers | Joleen Liang NEW ECONOMIES · 2026-04-15 Joleen Liang, co-founder of Squirrel AI, explains how its LAM engine—powered by over 20 billion learning behavior data points—replaces traditional teachers by handling 100% of knowledge transfer, turning instructors into data analysts and psychologists. She details how the Chinese Double Reduction policy in 2021 slashed revenue from $300 million to zero overnight, forcing a pivot to a hardware-integrated smart tablet and leading 80% of franchisees to convert $160 million in debt into company stock. The company now expands into the US with self-study centers and plans to introduce physical AI robot teachers within five years, while arguing that generic LLMs cannot deliver truly personalized education. Liang also shares how the co-founders maintained belief through the crisis, prioritizing health and exercise, and advises founders not to underestimate their abilities. ## Questions this episode answers ### Why does Squirrel AI use its own LAM instead of LLMs like ChatGPT for education? Joleen Liang explains that large language models (LLMs) make up less than 5% of their system and can't deeply personalize education. Their proprietary Learning Analysis Model (LAM) analyzes over 20 billion learning behavior data points — not just right/wrong answers but also mistake reasoning, time spent, and eye gazing — to diagnose why a student is stuck rather than just generating conversation. This allows truly adaptive, self-directed learning. [9:26](https://neweconomies.podhood.com/c22dbd90-77dc-4a90-95a3-52aca994ef5d?t=566000) ### How did Squirrel AI recover after China's Double Reduction policy wiped out its revenue? Joleen says the 2021 policy shut down their franchisees' tutoring centers overnight, sending revenue from about $300 million to zero. They pivoted from software-only to embedding the LAM engine into a smart learning tablet, enabling fully self-directed student use. Crucially, over 80% of franchisees converted $160 million in debt into company stock, believing in the AI's results, and many helped launch new centers under the compliant model. [26:19](https://neweconomies.podhood.com/c22dbd90-77dc-4a90-95a3-52aca994ef5d?t=1579000) ### How do students actually learn with Squirrel AI's system in a classroom? Joleen describes students sitting together in a room, each using a learning tablet with a headset. One instructor, acting as supervisor and data analyst, monitors up to 20 students via a dashboard that tracks progress and alerts them if a student is stuck. The instructor intervenes only to offer support, not to teach. Every student learns different content at their own pace, based on real-time AI diagnosis of their knowledge gaps and mastery levels. [10:13](https://neweconomies.podhood.com/c22dbd90-77dc-4a90-95a3-52aca994ef5d?t=613000) ## Key moments - **[0:00] Squirrel AI** - [0:00] “AI is replacing teachers. However, we do need teachers, but the roles of teachers should be totally changed.” - [0:30] When Squirrel AI launched, parents didn't want an AI system teaching their kids. - [1:02] Squirrel AI uses over 20 billion learning behavior data points for personalized content. - [1:57] Squirrel AI is a 12-year-old company building an AI virtual super-teacher for personalized learning. - [2:31] Squirrel AI embeds its AI engine in a smart learning tablet with optional offline self-study centers. - **[3:01] AI Speed** - [3:20] Squirrel AI was surprised by ChatGPT's rapid adoption but expected AI in education to go viral. - **[4:50] Role Shift** - [5:50] Squirrel AI supervisors act as data analysts and psychologists, intervening only when students struggle. - [6:30] AI-aided supervisors can be trained in one year, outperforming traditional teachers who need 10 years. - **[7:05] Future Vision** - [7:46] In 3–5 years education won’t change dramatically, but after 5 years AI will go viral. - [9:26] LLMs are less than 5% of Squirrel AI's system; they rely on their proprietary LAM for personalization. - **[9:54] Classroom** - [10:07] Q: How does a Squirrel AI classroom function for young students? A: Students learn on tablets, supervisors monitor and only intervene via alarms. - **[13:07] Global Adoption** - [16:28] US students are more active but Squirrel AI’s adaptive content helps them focus and learn. - [17:11] US third and fourth graders asked to continue learning after trying Squirrel AI’s adaptive content. - [17:23] Q: What do parents in China vs. the US think of AI-driven education? A: Chinese parents are willing to invest everything; US parents pay $500–600/month for results. - **[20:01] Data & AI** - [20:19] Joleen Liang calls the Chinese headband concentration experiment good but misunderstood by media. - [21:47] Squirrel AI tracks eye-gazing via camera for focus data, requires parental consent, and avoids cloud storage. - [23:15] Q: Why can’t LLMs provide personalized education? A: They lack analysis of learning behaviors; Squirrel AI’s LAM uses 20B data points. - **[26:02] The Pivot** - [26:27] China’s 2021 Double Reduction policy banned after-school tutoring, dropping Squirrel AI’s revenue to zero overnight. - [27:47] Squirrel AI pivoted from software-only to a smart learning tablet after the Double Reduction ban. - [29:52] The Double Reduction policy benefited Squirrel AI by eliminating teacher interference in student data. - [31:02] Squirrel AI had $300 million in annual revenue before the Double Reduction policy. - **[31:41] Co-founder Bond** - [32:46] 80% of Squirrel AI franchisees converted $160 million in debt into equity after the Double Reduction ban. - [33:57] Squirrel AI co-founders adopted fitness, healthy diet, and 7–8 hours of sleep to survive the crisis. - [35:35] Q: How do Squirrel AI co-founders resolve disagreements? A: Majority vote, with CEO Derek as tiebreaker; they accept wrong decisions together. - **[37:45] IPO Road** - [37:59] Squirrel AI operates with 300 employees generating around $300M in annual retail revenue. - [38:26] Squirrel AI targets 20 US centers this year, 50–100 next year, then expansion into Central and Southeast Asia. - [40:10] Q: Why is Squirrel AI pursuing an IPO despite being profitable? A: Long-term mission, investor demand, and global fundraising. - **[42:19] Robot Teachers** - [42:50] Squirrel AI will have a physical AI robot teacher with LAM brain within five years, serving as a hybrid mentor. - [44:21] “We want kids to focus on studying when they need to for one or two hours... Majority of your time, go out please.” - **[44:58] Final Word** - [44:58] Q: What is Joleen Liang’s biggest founder mistake? A: Underestimating her ability—she tells founders to believe in themselves. ## Speakers - **Olek** (host) - **Joleen Liang** (guest) ## Topics AI, Hardware & Robotics ## Mentioned Squirrel AI (company), AWS (product), ChatGPT (product), LAM (product) ## Transcript ### Squirrel AI **Joleen Liang** [0:00] AI is replacing teachers. However, we do need teachers, but the roles of teachers should be totally changed. We call them instructors, data analysts, supervisors—these three names actually are the same person, which means one person is in charge of the student to do monitoring during the student's learning process. So actually, in this whole AI era, we should have more previous teachers to be involved in AI education. **Olek** [0:28] What do the parents actually make of all of us? **Joleen Liang** [0:30] We have been in this market for 12 years. Basically, you can consider us as building an AI virtue super-teacher for the students. When we started to launch the product, it was very hard. The parents, they didn't really want an AI system to teach their kids. And now it's like, my kid's got to be involved with AI somehow. But not those AI chatbots, not those social media-generated contents to replace the human connection. That's not what we are doing. So our engine is called LAM. We've got over 20 billion learning behavior data to do analysis. So in that way, every student in the classroom is learning different content based on their knowledge state, their mastery levels, and everything. We want the kids to focus on studying when they need to. That's it. Majority of your time, go out please. We don't want them to stay here for a long time. **Olek** [1:24] What does the future of education actually look like? **Joleen Liang** [1:26] In the next 3 to 5 years, we're probably going to have. **Olek** [1:32] Joleen, we finally managed to make this happen. It's been a few months, but wow, you guys have been on such a journey, and an incredibly exciting one. So awesome to have you here, and welcome to the podcast show. **Joleen Liang** [1:42] Thank you, Oleh. It's my pleasure. Finally we met. **Olek** [1:46] Exactly. So Squirrel AI, maybe not many of our audience members know about Squirrel. What does it do, and why is the world of AI edtech so excitingright now? **Joleen Liang** [1:57] Yeah. We have been in this market for 12 years, so we are an old AI company already. So at the beginning, when we found this company, we started to build an AI engine to provide students with personalized learning. So basically, you can consider us as building an AI virtue super-teacher for the students, which means the AI can teach knowledge and abilities on behalf of the teachers to students, and it's very, very personalized, individualized, and tailored to each different student. And that's the software, the engine. And we embedded the engine into a smart learning tablet. And plus, even though everything is online,right, you can learn from the tablet anywhere. Plus is, we still have the physical, offline, self-study centers, which means the students can go there. There will be supervisors, instructors, and data analysts help with the students with during the learning and after learning, but not teaching knowledge. So that's what we have been doing for 12 years, yeah. ### AI Speed **Olek** [3:01] Wow. 12 years now. I mean, this whole AI category also has changed so much. As you say, you guys are slightly old,right? But ChatGPT is 1,200, 1,300 days old. How have you been surprised just how quickly things have progressed in the last three and a half years? **Joleen Liang** [3:20] Yes and no. Both yes and no. Yeah. When the ChatGPT was out and everyone was crazy, and that was something we didn't imagine. But what we did imagine is, it's going to go really, really viral. It's going to grow really fast. That's for sure. So when we started, no one knew what AI was, and people called it A1, and have no idea what AI was. When we started to sell the product and the service to the parents, they didn't really want an AI system to teach their kids. And now it's like the parents say, my kid's got to be involved with AI somehow. But not those AI chatbots, not those social media-generated contents. We want them to embrace AI in education, but they don't know how. So, you know, the desire, the drive is there already nowadays because of the fast growing of AI. So, yeah. **Olek** [4:21] You know, I was speaking to the CEO of Yahoo last night who was on the show, and we got to speak about tech trends. I actually brought up the trends around edtech because not enough people talk about it, yet it's such a big market. And you lightly touched upon that around all of these AI bots or AI tools that help the students at the end of the day. How do the parents and teachers actually feel about this? Is this going to replace them or help them? ### Role Shift **Joleen Liang** [4:50] From different aspects. Actually, in terms of teaching, teaching knowledge, training some abilities, and thinking methods, and model of thinking, and, you know, in terms of those teaching, the AI is replacing the teachers. However, we do need teachers, but the roles of teachers should be totally changed. That's why we always talk about upskilling for teachers or reskilling for teachers. So in Squirrel AI's ecosystem, the name, the title of teachers no longer exists. We call them instructors, data analysts, supervisors. These three names actually are the same person, which means one person, one instructor, one supervisor is in charge of the student to do the data analysis, to do monitoring during the student's learning process. And if the student has any struggle, no matter what the question could be, no matter what the scenario is, then the supervisor will need to talk to the student. So it's more like data analyst and psychologist and a friend and a mentor to be together with the students. Let's imagine how long do we need time and years to train a very, very good teacher. I would say 10 years. If we could have a very, very good teacher, that's already bravo. And now the teachers, they are trained in school, in college, five years, and they are not really qualified as a good teacher. But if we spent, say, one year to train an instructor or supervisor to be a good person to use the AI tool and help the students, and that person's going to be really, really professional experienced. So actually, in this whole AI era in education, we should have more previous teachers, instructors now to be involved in AI education. So yes, replacing partially, but we need more instructors and supervisors to be involved in education. ### Future Vision **Olek** [7:05] This is so fascinating. You know, when I also think about the education cycle for students,right, we spend, what, 15, 16 years in school. And if AI becomes that good, by the time we reach 18, part of me thinks, what happens to universities? What do these students go and do in the future? There is so much to kind of unpack here, but I think, like, one of my insights would be around, what does the future of education actually look like? Because I think so many people are confused. Is it just going to be students speaking to AI tutors with their instructors? Is that kind of what's going to happen? Maybe we're going to have AI universities at some point. I don't know. **Joleen Liang** [7:46] Well, yeah, I get asked this kind of question a lot. I would say in the five years, in the next three to five years, nothing will change dramatically. But after five years, it will probably go viral. However, I think the physical schools, either middle school, high school, university, are still going to be there. We still need physical interaction. Like when you reach out to me, I always wanted to sit in the same room together with you to meet. You know, we are human beings. We need to meet. We need to talk. But by using AI tools, it's going to be different, but the way of different, not only like interaction and conversation. So the AI for education, and probably for medical as well, is how AI can diagnose your needs, your demands, and then to recommend whatever that's the best and most adaptive, suitable for you to learn, to train, to, you know, talk. It's not like, I have a question, I ask AI to solve my question. That's very basic and simple. And if that comes to become kind of standard, it's going to be not really personalized. And people will feel really, really sad sometimes because the answer I always get is in the same format. It's in the same pattern. AI always try to please us. That's not what we really want. For once, twice, that's fine. So we need AI to fully understand us, especially in education. That's actually a different algorithm. That's actually a different AI structure, different engine. It's not large language models. Like LLM is only less than 5% in our system, and we don't really use LLM. We have our own LAM. So more, I think in the future, when people understand better about education using AI, they will need to build their own vertical AI education tools instead of just build on top of LLM to generate some conversation results. That's not AI future, actually. **Olek** [9:54] Yeah. And as we're all just sort of moving so quickly,right, what does this actually look like for the students then? So if you're a seven or eight-year-old sitting in a classroom and you're using Squirrel, how does that actually work? ### Classroom **Olek** [10:07] How do I know as a student I'm actually interacting with this AI companion? **Joleen Liang** [10:13] Great question. So the students will still be sitting in the classroom 5, 10, 20, 30, no matter how many students. They sit there, they use their learning tablet with a headset, and on the side, there will be an instructor or supervisor. And it could be me, Joleen, and I would just, I log into my data analyst platform, which means Joleen can look at 20 students at the same time. And Oleh is on this track. Someone is blah, blah, blah, blah. So that's called the learning process. And which learning objective you are learning, which video you are watching, how long have you stayed on this page. And if you stayed on this page, say, for five minutes, what you have been doing for five minutes. You know, if there's any concern, the system will pop up, say, alarm, go to see Oleh. And then I would just go to next week. Oleh, maybe you want to have a break. And are you struggling with this tablet? Or there's, you know, or maybe your headset. Something's going to be going on. So the instructor will be next to the students, but not teaching. And then after one session of learning, say, 45 minutes or one hour, and the instructor will guide them to do some activities, maybe go out, or maybe do some books, and maybe some incentives, some awards. Oh, who is doing great? Maybe some games. And then they can go back and start learning again. So in that way, every student in the classroom is learning different content based on their knowledge state, their mastery levels, and everything. So they learn differently, and the instructor will look at them. And after, say, one day learning, the instructor will have another very, very important role, which is data analysis. The instructor will look at, okay, there are 20 students learning the same domain or same scope, but Joleen has finished 10 out of 20. Oleh has finished 15 out of 20. And what's the mastery level of each learning component for different students? And then the instructor will understand the full picture of the classroom. And some students have mastered something the same. Some students have loopholes for different. Then the next class, the students still can learn based on their own pace. And the instructor can, you know, gather those information, and if necessary, they will talk to the students and to do a kind of conclusion and analysis. If necessary, they will need to talk to the parents. And sometimes it's not the students' problem. Yeah. **Olek** [12:50] It's so different,right, from when we were at school, now having all this, you know, AI companions and, you know, learning in a completely different way,right? It sounds like it's the future. I kind of almost see this, you know, we've both been in Asia for quite a long time now. I almost feel this is like a very Asian product. And I know you guys are now focusing on the US. What's the adoption been like in China? This is a Chinese company,right? Squirrel. What's been the adoption in China versus the likes of the US where you're now spending more of your time? ### Global Adoption **Joleen Liang** [13:25] Sure. Yeah. Our company was founded in Shanghai, China. However, when we started this company, our goal was not to, you know, build an engine that is only for Chinese students or Asian students. And our top scientists, they were recruited from the US at that time because we didn't really have that many AI scientists in the Asian market,right? So at the beginning, when we designed this AI engine, it's for international students, for all the students globally. So that's the goal. And because of this mission, this goal, when we designed the AI engine structure, everything should be, you mean, different cultural, different aspects need to be considered to build this AI engine. And after we launched the product in China, for sure, and then we realized, okay, this is not only for Chinese students, for sure, because all the students in Asia or Western countries, they have the same problems, same situations, and desire. And when we started to finally decide to, okay, we're going to launch to North America and other countries in the future, and we did market research in the US. And by the way, we are expanding our franchise self-study centers from this month in the US. So when we reach out to some parents or potential franchisees and schools, districts, the feedback is not as what we have heard outside. I think it's kind of the rumor. The students and the schools here and the parents, they still want their kids to grow faster, to have some kind of improvements. They are tired of the current school system here because everyone is sitting in the classroom learning the same thing, and the knowledge level is really low, lower than Asian countries. How can they catch up? How can their kids compete in the future? So when we bring this product to the US, of course, the engine is universal. We bring it to the US and place on AWS, the cloud here. And we also did a content localization. That said, we hired local American teachers and curriculum content designers. They are very experienced team members in the US and understand how US education systems work and the content. So we build those contents on top of the engine, which means the students here in the United States, they learn the contents from their own country and it's tailored to themselves and based on international universal engine. So I don't think there's a big, big, big difference, but the way that we're here is different. And also one slightly different is the students tend to be more active in the US. We are in Asia. They are more disciplined. That's one thing you can see straight away in the classroom. And that's good and bad because when the students are really hyped and excited, it's hard for them to focus. But once the AI engine can diagnose their loopholes and can be very, very adaptive to their current level, and plus if the contents are really interesting, then the student will start to focus and look at, and they will learn. They want to sit down and learn. So we've done some demos with third graders and fourth graders. They really, really love it. When we finished the session, they said, "Can we continue? We love to learn again." You know, that's a good sign. So students in the US, in Western countries, they still need some good learning system and contents, and they love. So there's no big difference. Yeah. **Olek** [17:23] You'reright on the discipline side. Having been in a European schooling system, discipline wasn't top of the agenda,right? But I also think a lot of the time is down to the personalization. You know, the teachers who we didn't enjoy, if we didn't enjoy the teachers who we enjoy learning from, you're not going to be disciplined. You're not going to be focused,right? So that's maybe, you know, one of the aspects where AI can help. What do the parents actually make of all of this? Do they really, are they a big fan? You're probably going to have a few which may question, "I don't like this." **Joleen Liang** [18:02] You mean in the US or in China? **Olek** [18:05] Both, China and US. **Joleen Liang** [18:07] I would say different stories. **Olek** [18:10] Really? **Joleen Liang** [18:10] Yeah. The culture is different. In China, actually, when we started, when we started to launch the product, it was very hard because no one knew what AI was. Like I mentioned, they didn't want AI to teach your kid. And now they really want their kids to sit down in front of the learning tablet, and there will be a good supervisor in guiding them. And the most part I love is the kids can learn by themselves because it's L5, autonomous, self-directed learning. They can focus, and they love learning. Once the parents say, "Okay, they love learning, they just want to pay for everything. They won't leave their kids." And the cultural differences in China, I believe in the whole Asia countries, is parents think education is the only way to get their kids out to compete. So they really want to probably sell everything and invest in education. That's what they want to do. And that's a totally different story in the US and Western countries. You know that,right? So where in the US, they're still willing to pay, say, $500 or $600 a month as long as their kids have some sort of improvements. Or their kids say, "Oh, bad mom. There's something I didn't learn from the school, but I learned from this center." And if they have this kind of feedback, they are happy to invest. Well, that's fair enough for us. We are not expecting them to spend $5,000 a month to us because AI can actually bring the equality,right, for the whole market. So that's good. So even though the cultures are different, but as I mentioned, the parents here in the US and in everywhere, they still want their kids to grow,right, in the way that they want. So yeah, I would say even though the culture is different, but the results, the facts are the same. **Olek** [20:01] Going back to this discipline point, I read a story. I think it was probably three or four years ago. This was in China in schools where these kids were busy having headbands,right, in the schools. And you'll know the story better, but they could basically track like how concentrated they were,right? ### Data & AI **Joleen Liang** [20:19] Yeah. Yeah. That story actually was a very, very good experiment. However, when I think Wall Street Journal, they interviewed, they wrote the story from another angle, which makes kind of pretty criticized, yeah, many media. I think on the way to AI education, it's a very good experiment. However, maybe people, the users don't really understand, and they are not comfortable of, you know, putting a device, pretty heavy device or light device on the head. But the data they censored is very, very important. It's the focus, the time of the concentration. And of course, there are different levels of the concentration. When we do something, especially we learn, like you spend one hour, I spend one hour. If your concentration level is nine, my one is five, who is more efficient, effective, for sure. So the data of the censoring is very important. So I would say, yes, that's a good experiment. But maybe in the future, in education, we're going to have some kind of smaller devices that won't have any negative effect for the students. But to censor their focus is very important. Like in our system, we have the multimodal of the learning behavior data. And actually, focus is one of the data. But we use the camera to capture the students' eye gazing and the movement. And if the, you know, the student is learning, reading, sorry, viewing a video, fast forward or backward, and for the five minutes, that's really, really important. However, if we do need the parents to sign the consent form, if they say, "Okay, no, I don't want my kid's, you know, facial expression to be captured," then they can choose to turn off the camera, but they will lose one kind of data, which means the AI system will not capture the eye gazing, the facial expression. Then they will be losing one kind of the data to do the analysis. This is actually important. And for us, when we do the live stream capturing the facial expression, we won't store them on the cloud because first, it's privacy. Second, if we record everything, that's too much storage,right? Anyway. **Olek** [22:41] For sure. **Joleen Liang** [22:42] So yeah, really depends. I think like in China, we have so many cameras, but it makes this whole society safe. So which do you want? There's nothing that you can have 100% correct. So if that can help the kids' learning and it won't have any harm and won't disclose your kids' private information, we just use this as a data to do analysis. Majority of the parents, they will say, "Yes, okay." **Olek** [23:15] Yeah. I'm not surprised. Also, I mean, you're just saying on so much valuable data,right? I can see so many different ways you could take this. You know, if I was a student using Squirrel, maybe only can have his own teacher, but you know, are you going to create your own LLM model? Are you using other models? Because you're sitting on so much data,right? And there's probably just so many opportunities where you can take this. **Joleen Liang** [23:40] Yeah. Well, if someone else is using some other LLM platform, they can't have the data, even though they build their own engine. Because LLM is based on very simple data. It's not much modality of the learning data. Where our engine is called LAM, the system records all the data to do analysis. So yeah, we've got over 20 billion learning behavior data. It's not onlyright or wrong. The time speed is not only about that. And for example, if a student is trying to answer a question and we are giving the same question, I have to use the pen to write on the pad, and then you do the thing, and the results, both of us get theright answers. But the AI will do that diagnose and analysis. Okay, Joleen, you know, had still two learning objectives missing, even though she got the answerright and only did everything correct. Then the AI, when they take this into consideration, the next step of the learning and the future steps of learning to me and to you are different. This is one, very simple one. And also we have the mistake reasoning analysis, like A, B, C, D. A is the correct answer. B, C, D are wrong answers. But B, C, D, wrong answers have different meanings, different reasons. And each different wrong answer has like five to ten backstory reasons. So if the student has one wrong answer, what is the reason? You know, there's all sorts of learning behavior there. The basic, the standard LLM cannot record this, cannot do the analysis, even though the basic LLM is like you ask questions and they give you the feedback. What they can record is only the history of the conversation, but they cannot deep dive into the core problems. They don't have that function. So that's why in education, we cannot use LLM as an education tool to teach students in the personalized way. **Olek** [25:46] It's so incredible. I know all this has really happened, like we said earlier,right? In really a short few years. I know you guys have been around for, you know, 10-ish years, but for the audience listening, Squirrel AI now, 50 million users, a couple hundred million dollars in revenue, give or take, a couple thousand learning centers. ### The Pivot **Olek** [26:06] But like all stories, and like all startups, there's always a terrifying moment. And one of your terrifying moments was back in 2021, I think. **Joleen Liang** [26:19] Yep. **Olek** [26:19] When the Chinese government decided to change their policy,right? This is the Double Reduction policy, where basically they banned all schools of having after-school extracurriculum because of putting pressure on kids. This must have been a pretty terrifying year. **Joleen Liang** [26:47] It was. It was the darkest time we ever had from 2021 to 2022. Well, luckily, because we have our AI engine, and first of all, actually, our headquarter, the company, we were not one of the banned company because we are still very supported by the government, AI-based company. However, at that time, we didn't have the smart learning tablet. We only had the software. Our franchisees at that time, so they had their after-school tutoring centers. They were using our software to give the students to sell to the parents. So they were banned. And one night, all of them closed down, and we had no revenue, zero revenue for probably almost one year. Yeah. And then at the end of 2021, we decided, okay, we're going to do a pivot. And of course, before the end of 2021, we did lots of preparations, and we actually had our very first smart learning tablet in, I think, in September, October. But the policy was announced in July. So it's just two, three months later, we had our very first version of the tablet. And we really appreciate our franchisees at that time. Even they were also very frustrated at the time. They had full belief in us, in the company, in the leader, Derek. And also they had full, full, full belief in the system. Because in the past, their kids, they've been using Squirrel AI system for many years already, and they got really, they entered good schools. And in their centers at that time, they had so many kids to use our system, and they got improved. And of course, they got money, you know? They had pretty good profit, good profit. So they actually encourage us to do the pivot. So we decided, okay, we need to have something to let the software to be embedded. Then that's the design of the smart learning tablet with the LAM system inside of the tablet. And when we found this company, when we started to design this engine, Derek started to name it as full self-directed learning by the student. He always said, okay, our system is a student-centered system. So actually, even though before the Double Reduction policy, our goal is to let the AI to teach the student 100%. But to be honest, before Double Reduction, it wasn't easy to let the students 100% to use the tablet. There are always teachers, they want to intervention, they want to do guiding. So the data actually was ruined because it's not the real data. Even though I answered the question correctly, maybe my teacher, my tutor kind of guiding me, and so I got this correct, but actually it's not correct. I didn't master. So actually, we need to say thank you to having this Double Reduction, which means no teachers could be involved, and we can finally come back to our full self-directed learning. So, you know, that's bad, but it was also good. We had a tablet, and then the policy encouraged the AI teaching. And because there's no teacher allowed, tutors allowed in the center. So yeah, so we did this pivot, and then we were back, our revenue got back in 2023. Yeah, I was back. So that's good. **Olek** [30:31] Wow. Okay. So the Double Reduction comes in '21. And did you guys know this was coming? Or was it, did you have a little bit of a heads up? Something is about to happen. **Joleen Liang** [30:44] We knew it was coming, but we didn't know that it was 100% cut off. Yeah. **Olek** [30:51] Okay. Wow. So you get a bit of a heads up. Basically, your revenue goes to zero overnight. What type of revenue are you doing before this Double Reduction comes in? **Joleen Liang** [31:02] I think the end user retail, you know, market revenue was about the same as what we had in 2023. It was probably 300 million US dollars at that time. **Olek** [31:15] Wow. **Joleen Liang** [31:15] 2023, it went back. **Olek** [31:18] So in '21, $300 million. **Joleen Liang** [31:20] Sorry, 2020. Yeah. **Olek** [31:24] 2020. And then as soon as the Double Reduction comes in, it goes to zero. Then you have to pick it up again. I read in one of your other interviews, you also had $160 million in debt. So this is not exactly, this is not exactly a fun ride,right? ### Co-founder Bond **Olek** [31:41] But I think in these moments, and they're quite special moments in a founder's journey because the ones who keep you going are your team, of course, but it was also your co-founder. And your co-founder, Derek, who's, from what I read, you know, an amazing AI leader in the edtech space alongside you. How did you two keep each other positive, but also, we have got this and we're going to figure it out? **Joleen Liang** [32:15] Sure. I think first of all is the full belief of AI in education. That's like four co-founders together, four of us and plus two partners. So the core team, four to six people, we had full, full, full belief. We know that AI for education is the future, no matter how hard it is. That's the first thing. The second support, I would say, from the distributors, or we call them franchisees. So they encourage us and they had full belief. They supported us. And for the debt, they own the debt. Now they are the shareholders of the company. So we paid about 20% in cash to some other people, to some other distributors, and 80%, more than 80%. They decided to transfer their debt into the shares in the company, which means they have full belief. And some of them, many of them, they started to open the new centers after, like from 2022. So yeah, they gave us great, great, great support. That's really amazing. **Olek** [33:21] Wow. That's amazing. **Joleen Liang** [33:22] We couldn't do anything. Yeah. **Olek** [33:24] Wow. That's amazing. So from your $160 million in debt, a lot of your learning centers believed in you, backed you, and instead of returning that cash, they converted it into stock. **Joleen Liang** [33:37] Exactly. So the second, yeah, from them. They are still, we are still very close friends. They have full belief in us. Of course, we fight sometimes about any product, but we love each other. They love this company. That's the second. Third, I think mental health and the real health. And we encourage each other to do training, exercise. So I was the first person to start going to gym, weightlifting. And Derek was the second person. And after I persuaded him for one year, he started. And then our CEO and another co-founder, CTO, David. So they didn't really want to go exercise. So we pushed them to go either running or walking or go to the gym. So four of us, we both in some kind of fitness. And when we had meeting before, I ordered kind of very healthy food salad, and they were eating, you know, very oily, salty food. And now when we have meeting, it's very easy. It's always salad, always salad, protein and good carbs. Yeah, I think we are very healthy. And we encourage each other to sleep early, to have at least seven to eight hours sleep. That's much more important than anything else. Even though it's, well, sometimes we do haveurgent projects, that's something else. But if there's a huge load of work, but we still encourage, I say, David, Jason, go to sleep and get up early tomorrow. We work tomorrow together. You know, that's good. So health. And do body check every month to make sure we are all good, healthy. That's third, but very, very fundamental. Yeah, important. Very, very important for us. But body health and also for mental health. **Olek** [35:28] I love that. It feels like a little family. **Joleen Liang** [35:30] Yeah, it is. **Olek** [35:32] How do you disagree with your family? **Joleen Liang** [35:37] How do I disagree with my family? You mean the. **Olek** [35:40] Yeah. **Joleen Liang** [35:40] You mean the co-founders? **Olek** [35:42] With the co-founders, yeah. So in every startup ride, co-founders are always going to disagree at some point. How do you navigate those challenges? And how do you actually learn to disagree with one another? **Joleen Liang** [35:55] That's interesting. Well, among four, I'm the only female. So three of them that men, they're strong, powerful. **Olek** [36:03] Amazing. **Joleen Liang** [36:04] Yeah. So if I have different opinion, I will just say it loudly and tell them the supporting points and the sequence might be, the risk might be. And we always have to kind of, we don't have to say, okay, Joleen has to beright or she has no option to say, to decide. If I'm in charge of this project or if I'm in charge of the North American, this market, then I'm the person. But if I am not 100% of what I'm doing, I always consult with them. They can give whatever the negative feedback to me, and I'm the person to decide if I am the person, you know, to be in charge. But if it's about the entire company's decision, so we always do, okay, we need to, I, Joleen, need to say if I agree or disagree. Everyone will say the answers. If it's two to two, like two people disagree, two people agree, then Derek will be the person to say yes or no. Yeah. But if three of us are saying, okay, we disagree on this project, and Joleen says yes, agree, then we follow the majority. So we have already set up the, you know, kind of the rules. Yeah. So it's easy. **Olek** [37:28] That's amazing. **Joleen Liang** [37:29] And we, yeah, we come to the same, like, comment that if we made the wrong decision together, it's accepted. **Olek** [37:41] That's good. That's a good leadership lesson. Okay, you guys have been going 12 years now, plus, under Squirrel. A couple hundred million dollars in revenue. What's the headcount now? Team size-wise? ### IPO Road **Joleen Liang** [37:56] Sorry, the. **Olek** [37:57] In team size? **Joleen Liang** [37:59] The team, oh, the team size, around 300 employees. Yeah. **Olek** [38:04] Wow. Okay. So 300 employees, a couple hundred million dollars in revenue. What does the future roadmap look like? A potential IPO maybe at some point? **Joleen Liang** [38:14] Yeah. For the revenue, for the end user retail market revenue, I think we have just about 300 million. Yeah, US dollars for the revenue. So for the future roadmap, yes, first step is the North American market, that's for sure. And hopefully we can have 20 to, you know, 20 centers this year and 50 to 100 centers next year, and then we can start to expand to other markets. So our next goal would be Asian, Asian countries, including Central Asia and also Southeast Asia. And we have actually received a huge amount of requests from different Asian countries so far. It's just we are only doing some negotiations, some conversations for now, but we won't really put into, you know, we won't be able to expand the market very soon in Asian markets, but it's huge, huge demand. But if they want to do some kind of collaboration, they want to place our AI engine into their local market and they have their own content or they want to use our US content or they want to use our Chinese content, we can do that straight away. So yeah, international markets, it's going to be the bigger picture for us. And we have been planning to do this, since we started this company. Derek said our goal is international market. As I mentioned, that's why we had so many scientists from the US, from even Europe. We want to be really, really an internationalized company. And for education, yeah, that should be universal. **Olek** [39:55] And then take it public. **Joleen Liang** [39:57] Yeah. We have already started to do preparation, but yeah, this year, next year, probably. It's not really in a debt rush, but it's a time. Yeah, we should. **Olek** [40:10] Why go public? You know, you see so many of these public companies going public and perhaps the performance isn't as anticipated. If the balance sheet is well and you guys are well funded, why not just keep it private? **Joleen Liang** [40:25] I think first of all, we do want to be a big giant company in the public in the future. That's the mission, the goal when we found this company. Second, we have so many investors and shareholders. I think they want it anyway. It doesn't mean that they want to sell their shares. They want to see their invested company to be public. It's kind of the owner. They want to keep it for a long time. And education is not a product you can sell and you get money back straight away. The majority of the people who are in our company as investors or shareholders, they have the passion about education. They want it for a long-term run. And if our company goes to global market, of course, they want our company to be public listed for sure. And of course, it will be very good for fundraising. We can attract different investors from everywhere. So yeah, it's good direction. **Olek** [41:29] It's one of the most exciting categories I think to be in. You know, when you said about expansion, especially in Asia, if you just think of the biggest Asian markets and similar cultures maybe to China, you know, Japan, Korea, India, even here in Thailand where I live, you know, AI isn't really like widely adopted yet. **Joleen Liang** [41:48] Yeah. **Olek** [41:48] But I imagine the population, where the countries are with the largest populations, that's probably where the opportunity is. **Joleen Liang** [41:57] You mean among the Asian countries? **Olek** [41:59] Yeah. **Joleen Liang** [42:02] I think India, Indonesia, very good. Vietnam is good too. Yeah. **Olek** [42:09] And Vietnam. Vietnam is super awesome. But come to Thailand. We need you in Thailand at some point. **Joleen Liang** [42:14] Yeah, Thailand. **Olek** [42:16] At some point. To finish this off, if I could, if you just look five years ahead, where do you think Squirrel would like to be? And then I'll come to my final question. ### Robot Teachers **Joleen Liang** [42:32] Okay. In terms of the global market, I believe we're going to be for sure in the North America, in the US, Canada, and probably one or two Asian countries, including Central Asia. And in terms of the product, we're probably going to have a physical AI robot teacher to be next to the student. But the key is not a robot itself, it's still the brain, which is the LAM engine. And we think each individual student and user, they should be able to have an AI virtual tutor next to them. It can be a physical robot. It can be on your phone. It can be on your laptop, anywhere together with you, not only teaching you the knowledge, but also to advise you as a mentor to be with you. Because that AI virtual tutor can be your closest friend to guide you. And also, of course, that mentor should be encouraging you to go offline to see other people, not to talk to me, always online. You know, it's got to be hybrid. It can't be just online only for anyone, students, kids, and adults. So yeah, I think AI virtual, maybe AI virtual mentor. **Olek** [43:57] Wow. If that happens, I mean, we've already seen this with robot chefs in the US. We're seeing a little bit with AI girlfriends, boyfriends. People will have their own opinions about that. That's fine. Teachers next. I mean, wow. If you guys make that reality, then the future of edtech and education is going to look completely different. **Joleen Liang** [44:21] Yeah. Well, the key difference is for boyfriend, girlfriend, I really don't encourage that at all. But it looks like the young generation, they want to use that kind of 80% to 100% to replace the human connection. That's not what we are doing. We want kids to focus on studying when they need to for one or two, three hours by using AI virtual tutor. That's it. Majority of your time, go out, please, with your friends, please, to the whole nature, please. We don't want them to stay here for a long time. But the key is they need to build their foundations, and then they will be able to grow faster in the future. ### Final Word **Olek** [44:58] Exactly. Exactly. Well, you've been through a lot in the last 10, 15 years, and I'm sure you speak to tons of other founders and coach them, you know, things you have learned. So for your Squirrel journey, what is the biggest mistake you have made which you coached other founders to hopefully avoid? **Joleen Liang** [45:21] I would say probably do not underestimate your ability. That's what I always did in the past. And I've trained myself mentally and always believe yourself that you can do something that is more challenging, that out of your expectation and out of other people's expectation. Always think you are the greatest person and try harder. Don't underestimate yourself. **Olek** [45:52] And you know, now with AI, anything is possible. **Joleen Liang** [45:56] Exactly. Yeah. **Olek** [45:59] Amazing. A beautiful note to end on. Joleen, this has been so much fun. Thank you so much for coming on the show. We can't wait to see where Squirrel goes next. And thank you so much for sharing your story. **Joleen Liang** [46:11] My pleasure. Great talking to you, Ollie. --- This library is powered by PodHood (https://podhood.com), the podcast website platform.