During the peak of COVID-19, the laboratory closed, and there was nowhere to go. The founding members from the same lab participated in an artificial intelligence (AI) competition. They turned the discomfort they felt during late nights into a business idea.
Coxwave is an AI startup that creates data analysis solutions needed in the era of generative artificial intelligence (AI). Founded in 2021 by four master's and doctoral students from the data mining lab at Seoul National University, they delved into the limitations of existing systems for data analysis while participating in AI competitions.
Coxwave initially introduced an image editing service called "Hama" and a generative image search named "EnterPix." However, it was difficult to analyze how users were utilizing AI with Hama and EnterPix. After selling these two services, Coxwave made its generative AI product analysis platform "Align" its flagship product and entered the business-to-business (B2B) market.
Align is a solution that analyzes various hints left by users in AI services, identifies problems with the service, and suggests improvement directions. Its distinctive feature is understanding context and intent, going beyond simple keyword analysis.
Kim Juwon (31), co-CEO of Coxwave, met with ChosunBiz at the office in Yeouido-dong, Yeongdeungpo-gu, Seoul, on the 26th of last month and explained, "If you say 'find it again' to a chatbot, it means that the results provided by the chatbot were not satisfactory, and you are seeking a more accurate answer. Existing services struggled to understand this subtle context. However, Align recognizes that the phrase 'find it again' indicates that the answer was insufficient and helps determine how to improve the clarity of the response."
Coxwave recently gained recognition in the domestic AI ecosystem through a collaboration with Anthropic. Coxwave is utilizing Anthropic's "Claude" model in its service. The fact that Align's service philosophy, which emphasizes "safety" and "reliability," aligns with Anthropic's direction helped facilitate the collaboration. Currently, Coxwave is supporting Anthropic's entry into Korea.
Coxwave, which attracted 4.5 billion won in investment from KB Investment, Danal, and Seoul National University Technology Holdings in January last year, has achieved a cumulative investment of 5.5 billion won. The company is currently collaborating with local businesses in India as a foothold for global expansion.
—What is the background of the creation of "Coxwave"?
We gathered in 2021 at Seoul National University's data mining lab to start a business. The founding members were one doctoral and three master's students, aged around 27 to 28 at the time. When COVID-19 closed the lab, we started participating in hackathons and AI competitions. Many competitions focused on tuning AI models to fit data provided by the organizers and checking accuracy rates. The process of conducting data analysis and model training repeatedly during such competitions became cumbersome. The idea for our initial startup was to create a solution that can analyze the data automatically and learn from it, merely by inputting the data.
—What kind of service is the main product "Align"? How does it differ from competitors?
Align is an AI product analysis platform that helps improve service performance based on user data. People express frustration or disappointment while using chatbots like ChatGPT and sometimes stop using them. We focused on the accumulation of negative experiences in generative AI services as data. Align extracts elements such as the reasons for user dissatisfaction in the dialogue data between users and AI, compresses it into around ten critical points, and delivers it to corporate clients.
The strength of Align is "personalization." Among the extracted data, there is also emotional data about users, so we reference many psychology papers and use various scales for emotion classification. Users provide explicit feedback like likes or dislikes but also leave hints about their preferences through various actions, such as pressing the copy button. We organize and provide this information in a way that AI product developers and managers can easily understand.
—Why did you transfer the two services, "Hama" and "EnterPix"?
Hama was an image editing service, and EnterPix was a generative image search service. Align started as a small analytical dashboard that we internally developed and used for the enhancement of Hama and EnterPix services.
Running these two services made us realize that operating generative AI products felt very different from existing web or app experiences. In web or app environments, the planner sets pre-defined buttons and input texts, leading to lower user freedom. However, generative AI products offer high user freedom, making it challenging to predict user behavior. It was not easy to analyze how users entered text, dragged to erase specific parts, or colored in Hama and EnterPix services. There was no analytical tool for generative AI products at the time, so what we internally developed and used formed the basis of Align.
While running the two services, we faced a crossroads: whether to continue focusing on image services as before or to provide B2B services based on the challenges we encountered. After much deliberation, we chose the B2B service direction and sold the two services to Sandoll when the opportunity arose.
—Currently targeting the Indian market. What is the reason for focusing on India?
India has a large population and diverse culture, allowing Coxwave to attempt many things. The country is also proactively adopting AI, leading to various startups, and the regulations are relatively lax. There are many Indian executives in the United States, which can help in expanding into the U.S. market, and being close to the Middle East and Saudi Arabia could also aid in expanding into the Asian market. We have signed a memorandum of understanding (MOU) with Renan, a data platform company in India, and are also collaborating with PwC India.
—Also aiming for entry into the U.S. Is there a specific timeline?
First, it is essential to focus on the Indian market to secure good case studies. To enter the U.S. market impactfully, it would be beneficial to achieve results in the large global market of India first, so we are operating on a two-track strategy targeting both the Indian and U.S. markets simultaneously. The transition to a co-CEO system is also aimed at conducting global expansion more efficiently.
We are also consistently conducting SaaS promotions and meetings with U.S. companies in the U.S. market. Our collaboration with Anthropic is part of our strategy for entering the U.S. market. We actively participate in developer-related events held in the United States, and at the end of last year, we also conducted Align advertising in Times Square, New York.
—What advice would you give to the government for the advancement of the domestic AI industry?
To secure global competitiveness, opportunities for going abroad are necessary. There is much support for early-stage startups, and we have received significant help ourselves. However, there is a vague period as we move beyond the initial stage. Especially in AI, it takes a long time to develop properly, and support tends to be concentrated in the early stages. It would be beneficial to have more diverse support when moving into the mature stage—not necessarily financial, but support that can assist in overseas expansion.