CEO Insights · Project X | Exclusive Interview with HUANG Yingning, Founder of Beipei Technology: When AI Becomes a Child's "Childhood Playmate", Should We Be Anxious or Embrace It?

Executives from Alibaba, Meituan, and Microsoft have all entered the arena, and Elon Musk has also launched an AI companion. This sector may seem booming, but behind it lies a common industry dilemma: the return rate is as high as 30% - 40%, product homogenization is severe, and "too much of a chat AI flavor" has become the most genuine feedback from users. Against this backdrop, the AI toy "Kedou Peipei" by Beipei Technology rocketed to the top 3 in the Tmall category within two weeks of its launch. What on earth did it do right? Facing the widespread homogenized competition, how can AI toys achieve true differentiated breakthroughs?

With these questions in mind, 36Kr's "CEO Insights · Project X" interviewed Huang Yingning, the founder of Beipei Technology, to uncover the secrets of creating a "growable" AI toy.

The main focus of this live - stream is on the following questions:

1. When did you start having the idea of starting a business? What was the most crucial factor that ultimately led to your decision?

2. It took less than a year from the company's establishment to the product's launch. How did you manage to achieve this so quickly? What trade - offs did you make at that time?

3. You raised funds three times within a year. What did the investors value the most? How did you feel when you received the first round of funding?

4. In terms of the technical approach, why did you choose to develop your own model instead of using those from big tech companies? Will there be a significant difference in the user experience?

5. Many parents are worried that overly intelligent toys will make children addicted to virtual interactions and reduce their connection with the real world. What's your take on this concern?

6. Looking back on the journey of creating the product from scratch, please share three of the most important tips.

The following is the conversation between the guest and 36Kr, with some content edited and organized:

36Kr: When did you start having the idea of starting a business? What was the most crucial factor that ultimately led to your decision?

Actually, at the end of 2022, about a month after the release of ChatGPT, the hype around it had reached its peak. I was in the postpartum period at that time. While breastfeeding my baby, I was chatting with ChatGPT, and it really blew me away.

I realized that the interaction paradigm between AI and humans had been completely revolutionized. Vaguely, I sensed that a new era had begun, and there would be a plethora of To - C opportunities. Every technological revolution gives birth to great products, especially those targeting consumers. And this time, it was a new technological revolution. As this feeling grew in my heart, I kept thinking. Moreover, my child was at the stage of starting to babble. As a working mother, I deeply felt that I didn't have enough time to accompany him, and the quality of communication between my family members and him also fell short of my expectations.

This made me start thinking: What could I do? As a new mother, I was recommended many parenting products on the market. At the same time, as an AI professional, I was also thinking about how to upgrade all these products. When I was doing my postgraduate studies at Peking University, I was engaged in interdisciplinary research on artificial intelligence and child psychology. We used AI to automatically identify whether children had early - stage autism tendencies and their attachment types. So I had an early exposure to child psychology, combined with my more than a decade of experience in the AI industry. All these were like a seed vaguely planted in my heart.

In early 2024, I took my child out to play. I heard a little boy asking his grandma a question about leaves, but the grandma said, "Don't ask these useless questions. Drink your milk first." I felt quite uncomfortable. I thought it was a great question, and we should popularize science and guide him to understand nature. I realized that there was no product on the market that could enable such interactions. With my background, resources, and team support, I knew I had to step out and make this happen. I had been thinking about it for a long time, but that moment was like a trigger, making me really want to create something to answer children's questions like that on the spot.

In child psychology, there is an important concept called "the first three years shape a lifetime." From a neuroscience perspective, this is somewhat true. The number of neural synapses in the human brain reaches its peak at the age of three and then gradually decreases. So, during this stage, how you shape a child, protect their curiosity, and guide their empathy will benefit them for a lifetime.

36Kr: It took less than a year from the company's establishment to the product's launch. How did you manage to achieve this so quickly? What trade - offs did you make at that time?

"Trade - offs" are very important. When designing the product, we were thinking: What is the core function of the camera? When interacting with children, what is the most important supporting function to create a "companion" feeling?

There is a famous "Rhesus monkey experiment" in child psychology. A baby monkey was placed with a cloth monkey and an iron monkey. The iron monkey had milk, while the cloth monkey didn't. There is an old Chinese saying, "One who provides milk is the mother," but it was not the case for the baby monkey. It would drink milk from the iron monkey and then lie on the cloth monkey. This shows that the preference for furry and soft things is deeply rooted in the mammalian instinct.

So we believe that such materials give children a sense of "life." Therefore, we chose plush over plastic and the current form over the camera. We think that high - quality voice interaction and interaction methods that match children's instincts, such as swiping cards and shaking, better meet their real needs.

Besides making many trade - offs in product design, our team also worked extremely hard. At that time, most of the core team members were parents of children aged 1 - 8. We were doing this with a "parent's heart." Everyone had a common thought: What kind of companion do I want to create for my child? When the first joint debugging was successful, our architect filmed a demonstration video of his daughter interacting with the product. Everyone said, "Brother Hao, you're happier than your daughter!"

Since the product was launched, in less than five months (about four months), we have upgraded it 41 times, maintaining an average of twice a week. This shows that we are a team that is extremely eager to iterate quickly and make the product better. We all hope that children will have more fun with it and parents will find it more valuable.

36Kr: You raised funds three times within a year. What did the investors value the most? How did you feel when you received the first round of funding?

I think there are two core factors.

First, the transformation brought about by AI, especially the upgrade of the toy industry and providing children with better growth and companionship experiences, is a highly promising and certain sector.

Second, our team is the "best fit" for this sector. I am both a mother and an AI professional. Our team has a complete range of talents in various functions, and everyone joined because of the common vision. In terms of the starting point and dedication, we are the best fit for this cause.

I once asked the first investor who invested in me, "Why did you invest in me at that time?" He replied, "You gave up the Alibaba P10 position twice to start a business. I think 'resilience' is very important in this sector. I believe you will succeed, so I invested in you." I found this quite interesting. Later, on Thanksgiving Day that year, I received a gift from that investor - a small keychain that said, "Accompany you to believe in the impossible." At that moment, I was really touched.

As humans, before achieving something, it's hard to imagine what the future paradigm, parenting model, or consumption model will be like. But once it's done, it all makes sense. So I think "willpower" is very important. It's crucial that you are willing to believe in what the future world will be like and work hard for it. What the first - round investors valued was actually the strong willpower of our team, including myself.

When I actually received the financing, I was quite calm and didn't have many intense emotions. Then I flew to Guangzhou because we have a supply - chain team there. I had a small team - building activity with them, had some drinks together, and reassured the team so that they could focus on product refinement and market promotion without distractions. That's when I was really happy.

36Kr: In terms of the technical approach, why did you choose to develop your own model instead of using those from big tech companies? Will there be a significant difference in the user experience?

I don't think we are in competition with big tech companies. Big tech companies are like "water," and we are like "boats." When the water level rises, the boats will float higher. For AI application entrepreneurs, the opportunity always lies in how to make good use of all the infrastructure, including the basic models and tools provided by big tech companies. When we say we "didn't use the models from big tech companies," it means we didn't simply call their APIs to create a simple dialogue system. In fact, we conducted a lot of engineering optimizations based on multiple models, including some SFT and subsequent post - training, all of which were deeply optimized according to our specific scenarios.

For a startup, first of all, this is affordable because our team has strong AI capabilities. Our chief scientist and others are capable of "polishing these tools to be more refined and useful." Second, if you don't do this and don't take the initiative to "sharpen the knife," the evolution speed of the overall product experience will slow down. Many times, the optimization points for product details are marginal data in the data - set distribution of big tech companies, but they are essential for us. And once we do it well, it becomes our competitive advantage.

When I was the CTO and wrote algorithm codes myself, I always believed that all technical decisions and architectural designs must serve the business. It's the same now. Whether we choose to train or not, use or not, the core is to serve the business. Currently, many aspects require a lot of engineering optimization. Big tech companies provide the infrastructure. For example, big tech companies may be like "power generators," and if you want to use electricity, you may need transformers and other equipment to adjust the voltage to a suitable level for you. That's your job.

But I must emphasize that we don't develop base models. We have a feedback loop of proprietary data based on our own scenarios and context. Therefore, we can be more refined in a very narrow and vertical field and consider children's needs more carefully. Their products are general - purpose, and there is no conflict between us. If their capabilities improve, it means my base capabilities also increase, and the optimizations I make on top of it will further enhance the overall intelligence level.

Interacting with children cannot be solved by a single model. We have deployed many "Agents" on the rabbit and bear toys. For example, when a child wants to play a sound - guessing game, it's one Agent; when they want to sing, it's another Agent; when they want to play interactive stories or even check the weather, they are all different Agents. A general - purpose model may not be able to handle all these business details. But for children, their needs are diverse. We have a set of "gamified content," such as various educational cards; "narrative stories," which are about how to play interactive stories with children; and "emotional resonance," which focuses on chat and emotional companionship. Chatting is just one part of our product. In addition, in our model, we specifically adjusted the response style for children. You may notice that some products on the market that interact with children give very long responses. When a child says just a little, it replies with a long paragraph, which is quite overwhelming in terms of language. Children can't even listen to it all. So we made some adjustments at the model level according to children's characteristics.

36Kr: Many parents are worried that overly intelligent toys will make children addicted to virtual interactions and reduce their connection with the real world. What's your take on this concern?

Our design concept has been very clear from the beginning: how to help children better connect with the world. The physical interaction, physical cards, parent - child topics we designed, and teaching children how to use the toy to make their friends happy - all these stem from one intention: Our Xiaokedou hopes to accompany you when you have no one around and help you have more fun with others when you do.

On the other hand, I also think that AI is an irresistible trend. So how can we integrate our parenting concepts and worldviews into AI, present them through it, and at the same time, let parents enjoy the convenience it brings? What does this convenience mean? For example, after a child interacts with it, parents can better understand their child.

Some time ago, we developed a preliminary "mental model." By analyzing the conversation data between my child and the product, we were able to output various characteristics, potential, and talents of the child. I think the analysis result was quite accurate. As someone who studied child psychology, I usually observe my child. I feel that she is not a "straight - A student." She is not very interested in reading books, but she has a very high emotional intelligence and strong empathy. For example, she would say to Xiaokedou, "Let's give mom a surprise, okay?" When I came back, she really gave me a diamond sticker.

She is developing her own growth path. I believe that in the future, she will have a better life because she is good at providing emotional value to others and has strong empathy. The other day, she was sick and didn't go to school. When she saw the security guard, she said, "Uncle Security, hug me. You haven't hugged me for two days." So I think she is a child with unique talents. But many parents may not have the time and professional background like me to make such insights. And "Kedou Peipei" can present such insights to parents, telling them, "What are the characteristics of your child?"

When a child is born, they have an initial "schema," which can be understood as the innate structure of genes and the brain. I once watched a documentary about a climber who was determined to climb a nearly 90 - degree cliff without any protective equipment. Someone asked him, "What's the point of not wearing protective equipment?" He countered, "Then what's the point of wearing them?" Later, scientists scanned his brain and found that his "hypothalamus" was extremely insensitive. This means that his feeling of walking on the cliff was the same as our feeling of walking on flat ground. That's the climber's talent.

The brain has its original structure, which determines what we are sensitive to, interested in, and absorb more quickly. If we can discover these characteristics of children earlier, we can magnify their strengths infinitely. Moreover, the future society will be very diverse. Currently, parents generally have two pain points: First, they don't have enough time to observe their children; second, they lack professional knowledge. Even if I see you doing something every day, I don't know what it means and where your talents lie. So we hope to make more contributions. This is similar to my work in the Peking University laboratory before: using various devices to capture children's information and then telling parents whether their children have early - stage autism tendencies and their attachment types. We also hope to provide parents with more insights into their children's growth.

36Kr: Looking back on the journey of creating the product from scratch, please share three of the most important tips.

First and foremost, it's the scenario. What fundamental changes does your project bring with or without you, with or without AI? This fundamental change determines the ceiling of your project. For example, we asked ourselves: Why are we making an AI toy? Because we believe that people in each era have their own opportunities, and the opportunity in this era is AI. If this project can be realized without AI, I'm sure there are smarter predecessors who would have already achieved it. So AI is our opportunity. But the key is what unique value can these AI - based things bring to users?

Second, be patient and respect the laws of nature, just like "sowing in spring and harvesting in autumn." Even a flower or a fruit has a natural process of growing, ripening, and becoming sweet. Forcing growth is often not a good thing.

Third, iterate quickly. From day one, think about the connection with users and iterate quickly. Act quickly, but don't be overly impatient with your goals.

For more exciting content, click to watch the full live - stream replay: