SEOUL — The day will soon come when artificial intelligence provides a full-length explanation for all math problems and even inspire the students to do better, at least according Lee Yong-jae and Lee Jong-heun, the founding CEOs of South Korea’s up-and-coming educational technology startup Mathpresso.
Mathpresso is a startup that operates AI-powered edtech platform Qanda, through which the company raised a total of around 70 billion won of investment since its establishment five years ago.
Mathpresso, led by the two 28-year-old CEOs, now has more than 100 employees working at its headquarters located in southern Seoul.
The ambitious venture began in 2015 when Lee Jong-heun, then a private math tutor, met his high school friend Lee Yong-jae, who was a computer engineering major looking for a business idea.
“I was teaching at a private educational institute. Back then, many students would take photographs of their math problems and send them to me to see if I could help solve them,” Lee Jong-heun recalled. “I thought, why not create an online math Q&A platform?”
The two soon set out to work, although the beginning was humble.
The company started out with a simple platform operated by a small group of college graduates.
The platform served as a matchmaker between students and tutors. The idea was to bring private tutoring sessions online.
But while the idea worked, the process was time consuming.
They would receive around 500-1,000 questions a day, and most of them were handled manually.
Since the platform did not store a database of the questions and answers, even if a different student asked the same question, another tutor would have to be assigned.
Then in 2017, the company developed its own search engine that can recognize text and mathematical symbols from photos and store both the questions and solutions provided by tutors.
“Most image search engines back then only processed natural languages, like English or Korean. As they were not able to scan mathematical symbols, we had to develop an engine on our own,” Lee Yong-jae said.
Backed by its new AI engines running on the Qanda platform, Mathpresso is now able to automatically sift through questions and pull out solutions that are saved in its database.
Qanda is basically an online application that uses AI engines, which recognizes math problems that users upload as photos. The AI engine then goes over its database to see if there is an explanation for the problem. If not, the platform assigns a tutor to provide a full-length solution in minutes or solve them on its own if they are simple enough.
Mathpresso uses Amazon Web Services’ graphics processing units as its hardware, and the startup is one of AWS’ biggest clients in Korea, according to the CEOs.
Qanda now can handle a much greater number of math problems at a faster pace, and not just from within Korea.
Qaunda receives 3 million questions daily from different nations, including Korea, Japan, Vietnam, India and Indonesia. Around 99 percent of questions are answered by the company‘s artificial intelligence, and the remaining 1 percent are manually handled by tutors.
In South Korea alone, Qanda has so far recorded around 5 million downloads, equal to two-thirds of the school age population — elementary and secondary school combined. It says it has around 1.5 million monthly active users.
“We figured one of the important things in operating the service is to provide solutions as fast as possible, say in five seconds for instance. To do that, we needed enough data compiled in our database and substantial amount of users who would continue to ask questions that would pile up in our databank,” Lee Jong-huen said.
More data will provide bigger data sets to train its machine engines and increase the accuracy of the model, and that is how Mathpresso aims to seize the edtech market here.
Of all nations that Martpresso provides its service, Korea is the most matured market in terms of the number of users and questions asked, they said.
Korea will naturally become the first test bed where the company will try out a billing model, the CEOs said, saying that a billing model will be available sometime next year. Currently, Qanda allows students to search math problems and solutions for free. Students only pay per use when the questions that they bring are not in its database and thus have to be matched with human tutors.
Mathpresso also plans to launch in to more countries.
“We are planning on launching the service in other nations, such as Taiwan, US and some South American nations,” said Lee Yong-jae. Lee added that they have sorted out the candidates based on the average hours that students spend on studying mathematics.
When expanding abroad, Mathpresso could face tough competitors.
China’s internet firms, such as Tencent, Alibaba and Baidu, are big players that offer similar platforms, US-based firms like Google and Microsoft, meanwhile, tend to use the technologies in more research oriented terms, the CEOs added.
To win the markets here and abroad, Mathpresso will prepare for adding new features to Qanda, they said.
“There are three reasons why people want have good tutors. First, outstanding teachers have quality study materials. Secondly, they quickly understand strengths and weaknesses of their students. Lastly, they know how to motivate them. We would like to bring all these aspects to our platform,” Lee Jong-heun said.
Mathpresso will open up a market space where users and partners can upload study materials. Mathpresso also plans to collaborate with both the public and private education sectors to secure a wide range of study materials and enhance its AI-powered recommendation engine. The company’s recommendation engine will be capable of analyzing user data, evaluate the strengths and weaknesses of each student and tailor study materials it offers them, the two CEOs said.
“We can add gamification features to the platform. The platform can provide a math question to student A, saying ‘The problem was previously asked by your classmate, and you might want to give it a try,’” Lee Jong-heun said, explaining one option for keeping users motivated.
When asked how fast the edtech industry and related technologies will grow, Lee Jong-heun said, “It will depend on how much resources key players invest. The machine translation improved quickly based on users online data and searches. YouTube was able to improve its video recommendation engines based on numerous users and searches on the platform.”
About the prospect of Mathpresso, Lee Yong-jae said, “We both think Mathpresso has laid building blocks for a platform that will constantly secure data that students may ask.”
Lee Jong-heun added, “It’s not like we will be able to outperform Google’s technology. However, we think Mathpresso has the upper hand at least when speaking of such a platform.”