Korean: 
프랑스 비바테크 2023 인터뷰와 인사이트 정리
프랑스 비바테크 2023 인터뷰와 인사이트 정리
This is an article summarizing my experience at Vivatech, one of the largest IT/startup events in Europe. The article focuses on the problems, items, and insights gathered from various companies participating in the event, based on their names and nationalities. The price for a 4-day pass to the event is approximately 600 euros.
Presentation by a startup that creates airplanes capable of traveling from Paris to New York in just one hour
In this article, I will cover Korean companies more extensively for two reasons. Firstly, Korea was selected as the country of the year at Vivatech. As a result, there were many Korean companies to explore. Secondly, I am proficient in the Korean language, which naturally allowed me to gather a higher quantity and quality of insights.
Exclusive presentation space for Korean startups
Some companies didn't engage in extensive conversations, and I guess there are a few reasons for that. Firstly, it could be due to my limited English skills, which prevented me from fully understanding everything the other party was saying. Secondly, some people were so tired from the long event that they simply didn't want to engage in conversations. Lastly, there were cases where initial interest faded as I approached to inquire and investigate further.
Through the interview, I made efforts to uncover the stories behind the initial startup journey and the hidden aspects behind the product itself, such as how they discovered and tested the item to reach now, and how they found the fit between the market and the product (PMF).
Petnow
Only a half(53%) of animals was registered to Animal Identification System
Explained by CEO Lim Jun-ho.
•
Problem the Product Solves:
1.
The South Korea government introduced the Animal Identification System to address the issues of stray and lost pets. However, the registration rate remains low.
2.
Inserting chips into pets' bodies to enable pet identification was suggested, but many owners are hesitant about this invasive method.
3.
While attaching external chips to dog collars is the most common approach nowadays, it has limitations due to the method itself. That is ‘external’ solution.
•
Product:
◦
They are developing an application that utilizes the protruding nose for dogs and the face for cats to identify their identities.
◦
It's worth noting that they attended CES. Although many companies got CES Innovation Awards, very few receive the CES “Best” Innovation Award. Among Korean startup companies, excluding major companies, Petnow is the only one to have received this prestigious award.
•
Other Insights:
◦
Q: Why does the product use the nose for dogs and the face for cats for identification? A: It is because cats often have fur covering their noses, making it challenging to use the nose as a reliable identifier.
◦
Q: It seems to be based on artificial intelligence. How did you collect the initial data? A: We went around and captured images of each individual dog and cat on random places.
◦
Q: I'm curious about the business model. A: We are currently collaborating with pet insurance companies. Additionally, we believe it can also be used to find the owners of stray animals when they are discovered.
There were likely technical challenges, but I was amazed by the fact that they were able to solve many problems simply by utilizing the folds on a dog's nose, which I had previously overlooked. It's an incredibly intuitive product, which makes it even more impressive. The demo was also very cleverly executed. Since they couldn't show the process of capturing a dog's nose in real-time (as they cannot bring a dog!), they brought two dog and cat models for each and demonstrated how the system could distinguish between them.
MOREH
It was a conversation with (Co-)CEO Yoon Do-yeon.
•
Problem the product aims to solve:
◦
Recent machine learning models are highly diverse and complex. However, the parallelization technology that supports the GPUs used for machine learning model training is inadequate, as a single GPU is often insufficient.
◦
NVIDIA's GPGPU, driven by CUDA, has been widely used and practically monopolized the market for machine learning tasks. As a result, purchasing and using NVIDIA GPUs for machine learning is expensive.
▪
Note: CUDA is a library and compiler set for executing user-defined programs on NVIDIA-based GPUs.
•
Trigger for discovering the team's capabilities and commercialization potential:
◦
The team originated from the Supercomputer Research Lab at Seoul National University. They have been building GPU-based supercomputers since 2010.
◦
Despite having limited research lab resources at the time, they participated in a supercomputer competition and achieved an outstanding ranking of 255th globally, which highlighted their potential.
•
Product:
This section contains technical details.
1.
The team is developing the MOREH abstraction layer, which allows combining N GPUs to function as a single virtual GPU. This abstraction layer handles various forms of model parallelism or data parallelism automatically, relieving developers from the burden of considering multiple parallelizations when writing source code. Developers can write their source code with the mindset of utilizing a single high-performance GPU.
•
Note: Model parallelism involves distributing the model across multiple GPUs, while data parallelism involves distributing only the data across multiple GPUs, synchronizing the parameters appropriately. Writing source code to consider multiple parallelizations on multiple GPUs is a tiring and challenging task.
2.
The MOREH abstraction layer can be divided into two main components: the frontend and the backend. The frontend generates a graph based on PyTorch or TensorFlow's eager mode. The backend optimizes the graph generated by the frontend.
3.
The MOREH abstraction layer is not dependent on NVIDIA GPUs. To break this dependency, they developed an abstraction layer called MoDNN on top of OpenCL. Currently, the focus is on development for AMD GPUs, but there is also the possibility of Intel producing GPUs, as well as utilizing other machine learning-specific compute devices. Hence, they should not be tied to specific platforms, including not only CUDA but also AMD's offerings like ROCM.
•
Insights:
◦
Q: Is it available to try the service? A: Yes, it is available. The service is accessible through KT Cloud under the name "HyperScale AI."
◦
Q: I understand that TensorFlow draws the computation graph in the backend. Is it the same for PyTorch? A: It is correct that we draw the computation graph for PyTorch as well. TensorFlow does draw the computation graph, but we are better at it. While TensorFlow focuses on optimizing the execution of specific operations through "local optimization," we draw the graph for "global optimization" of the entire graph. Therefore, TensorFlow can also be executed in eager mode.
◦
Q: I heard that models can be optimized using high-level APIs like TensorRT. What is the difference in this regard? A: CUDA was created a long time ago and carries the burden of legacy. Consequently, there is not much room left for optimization. Hence, they attempt to provide various software on top of CUDA, but it doesn't work as well as expected.
◦
Recently, PyTorch has started supporting tasks related to parallel computing. Unfortunately, our solution is far superior, so if a PyTorch user has written parallel code based on PyTorch, we need to remove all the parallelization code from the backend and repeat the process of creating the computation graph. That's why we have been in contact with the PyTorch team, suggesting they simply use our solution... (I didn't fully understand it)
◦
I believe there are many attempts to create products and solutions based on machine learning at a high level of abstraction especially in Korea. However, companies approaching the system level like us seem to be relatively rare. We have accumulated expertise by focusing on GPUs for a long time, and we believe we possess technology that is on par with leading global companies.
I felt that this technology is genuinely necessary. It made me think that this is what a "software tech startup" should be. If I weren't already considering entrepreneurship, I would have wanted to apply for an internship. Companies that build such deep stacks vertically must have put tremendous effort into understanding what "proper abstraction" means. I believe we can learn a wealth of knowledge from them, comparable to any major corporation.
Bryanthings
It was a conversation with an anonymous project manager.
•
Product:
◦
Create anything that involves personalized products in the retail market.
•
Insights:
◦
Q: The description on the signboard doesn't provide a clear understanding of what the startup does. Could you explain it more? A: We are a company that has been in operation for 9 years, so we can no longer be considered as a startup. We specialize in building personalized experiences in the retail market. It's hard to pinpoint one specific thing we do because we continuously produce such diverse products. (Showing a kiosk with an iPad attached to a printer) For example, if you need a ribbon to wrap a gift box and you want a personalized message printed on it, we can fulfill that demand. Another example is our collaboration with Dior. (Showing a video) We built a setup where various perfumes are displayed, and as customers answer a series of questions, only the compartments with recommended perfumes light up.
Example of providing individualized results
◦
Q: I'm curious about how companies that request such work approach you for collaboration. A: Some come with specific ideas, saying, "I want to create something like this, can you make it?" Other times, they come without a specific idea, and together we build and commercialize personalized experiences. It's possible because we have the know-how to build such things.
◦
Q: If you have to create each item in a customized way, it seems like the cost of producing these personalized devices would be quite high. A: It's not necessarily the case. If you look closely, you'll see that it's just pieces of wood put together, and it's not difficult to do this or that... (The misconception that expensive materials were used and extensive effort was involved disappears.)
The theme of personalized experiences, in reality, is dominated by "AI-based personalized experiences." It wouldn't be an exaggeration to say that it accounts for 90% of the significance for me. It's astonishing that a company can be created, generate revenue, and sell such experiences with just planning, design, and a bit of tech.
Instagram (@bryanthingsparis)
Instagram photos and videos
https://www.instagram.com/bryanthingsparis/
I'm attaching it in case it can inspire my friends who work in design.
RC Robotics Cyborg ️
•
The reason I glanced at this company's booth, despite not being particularly interested, was that I wondered why they had placed such a massive robot in the section for tourism booths. I couldn't contain my curiosity and asked, "Why was it tagged as tourism?" They explained that the event organizers didn't have a robotics section, and since the robots they sell are related to the restaurant industry, they were tagged under tourism. So, I inquired about what the company does, and it turns out they are not actually manufacturing robots but acting as intermediaries in supplying them.
•
Insights:
◦
Q: How did you get started in this type of business? A: (I didn't fully understand) Initially, they planned to operate a restaurant and intended to hire robots. However, they ended up abandoning the restaurant plan and delved into the business of supplying robots.
◦
Q: How do you sell the robots? A: We use various methods that are well-known, such as visiting stores for sales purposes and conducting promotional events.
◦
Q: When you look at French restaurants, the spacing between tables is usually quite narrow. It seems like the robots wouldn't be able to navigate through those spaces. Do you still think there is sufficient demand? A: Well, in Paris, that's true, but if you go to the suburbs, it's not as cramped.
◦
Q: Do you only sell one type of robot? A: No, we sell serving robots like this, robots with trays, and we also sell this robot and that robot.
Once again, it was a moment that made me realize how diverse the business world is. As the market for serving robots began to grow, companies emerged that supply products centered around the theme of robotics. I'm curious to know what could be the main difference when other robot suppliers enter the scene.
N.LIGHT ️
The CEO, Park Jin-young, and 3D designer, Kim Seo-yoon, kindly met with me.
•
Problem the product aims to solve:
◦
Let's imagine a company that manufactures hardware products like smartphones, and their designer creates 3D rendered images. Rendering complex 3D objects can take days. However, the person providing feedback on these images is a non-expert who lacks knowledge of 3D design and rendering tools. In such a situation, even if they want to make a simple change, such as altering the color or curvature of the smartphone's edges, they have no means to do so. Additionally, there's the issue of needing expensive 3D software just to open the rendering files.
•
Product solving the problem:
◦
Our product enables real-time collaboration with 3D objects, similar to multiple people working together on a Google Doc. Since it operates on the web, it allows for easy sharing and opening of design files without the need to install separate software. The user interface is designed to be user-friendly even for non-experts.
•
Process of discovering the problem:
◦
Q: How did you come across this item, and could you tell us about your background, CEO? A: I was originally a developer, and our item didn't involve developing the product after finding product-market fit (PMF). Our team's first item was software related to 3D printers. However, as the 3D printer market wasn't as large as expected, I went against the opposition of my team members and decided to pivot. Our second item was 3D rendering software integrated with VR. Since we were developing this item before the VR boom took off, we were able to ride the wave of VR. During that time, someone I met suggested, "What if you try doing it on the web?" We made another pivot before the popularity of VR waned and transitioned to our current item. Despite the existence of easy-to-use 2D image design tools, we observed significant growth in software like Canva that focused on collaboration and easy sharing. Our team was fortunate that everything fell into place. If we had started this item five years ago, we might not have succeeded. This is why I believe luck plays a significant role.
•
Insights:
1.
Upon reviewing the companies present here, investors noticed many Korean startups are focused on "deep tech" fields, while there are many French companies that emphasize a "soft and cozy" lifestyle. They suggested that starting a business here could be a valuable experience.
2.
Q: As someone with limited experience, I don't have significant technical skills. Therefore, I tend to consider service startups, but I worry that the depth of my ideas might be too shallow. A: To create a service like Baedal Minjok (food delivery service), you don't necessarily need advanced technical skills. You can still be successful. However, I believe that technology startups have a distinct advantage. They are better equipped to maintain a lead that is difficult for latecomers to catch up with. I'm not saying that starting a service startup is bad. It's important to think about how to outrun the latecomers. Also, it's helpful for the CEO to have a certain level of knowledge in development when working with developers. Nowadays, developers are individuals who possess the ability to create the products we imagine. I think it's a waste for CEOs not to leverage their abilities in development and start a business solely as developers.
3.
Q: I'm not sure if events like this are really helpful for your company. And I worry that if students like me keep asking questions, it might become a hindrance rather than a benefit. A: We recently received funding, so we're not in desperate need of it. One of the purposes of participating here was to find partners, and compared to CES, it seems like there is a higher proportion of students here rather than company officials or partners. Asking questions is not a problem.
4.
When I visited them again on the last day of the event, I could see the CEO and employees engaging in more lighthearted conversations and enjoying themselves. I learned that their product will be used as educational software in schools and heard about the disconnection between this 3D design software and the process of manufacturing goods in actual factories. The CEO also mentioned that while many students were looking for internships, few asked about these aspects of the business. He added that he believes the business will be successful.
Complicated things often become complex for a reason. However, that reason doesn't always remain important. Even if you keep control over the details, there are people who seek abstraction and simplicity.
La winetech
Sophie Blumm (not certain) greeted us warmly. She spent the whole day talking, but despite not being a customer, she was eager to explain everything to us.
•
Problem to be solved by the product:
1.
Leftover wine management issue: For example, when a 750ml bottle of wine is opened and only 150ml is sold, the remaining 600ml is left exposed, and it's uncertain how it will be stored. This poses a risk for someone who might end up drinking low-quality wine.
2.
Problem of lengthy serving time: It's challenging to pour the right amount of wine according to the customer's order. Estimating or measuring the quantity while holding a wine glass can be quite difficult.
3.
Variation in optimal serving temperature for different wines: Wine tastes significantly different depending on the serving temperature. Just as no one wants to drink warm soju, different types of wine, such as white, rosé, and red, have different average serving temperatures. Even among red wines, the recommended serving temperature can vary based on grape variety or specific wine.
4.
External events: There are often occasions where wine needs to be served at external events. When transporting wine in bottles, it is exposed to light and heat, making it susceptible to damage, and it generates a considerable amount of waste.
•
Solution provided by the product:
1.
They created a machine in the form of a water dispenser that can accommodate a reusable 2-liter plastic container. This container completely seals the wine, allowing it to maintain its quality for 4 to 6 weeks.
2.
The plastic container is equipped with RFID technology. When the container is inserted into the machine, the wine is cooled to the appropriate temperature for that particular wine.
3.
The machine has a one-touch button interface, allowing servers to choose between pouring 3.8cl, 12.5cl, or 25cl with ease, similar to a dispenser at McDonald's. Servers simply place a wine glass under the dispenser and press the button to fill it.
•
Curiosities about the product:
◦
The water dispenser-style machine costs around €100 per month (approximately KRW 140,000), including delivery. If the 2-liter container is filled with wine, the price can range from around €18 (approximately KRW 25,000) without delivery, depending on the wine type. Customers borrow the machine and return the empty container when they request more wine, receiving a new one in return.
◦
The wine producers do not handle the task of filling the containers. The company takes care of transferring the wine into their dedicated containers. There is nothing particularly special about sourcing the wine from the producers; they simply bring in the wine when it is requested, usually in containers holding several hundred liters.
◦
There seems to be some additional functionality. Customer preferences and consumption details, such as their preferred wines and the amount consumed, are tracked through a separate application.
◦
The company originated in Belgium and currently has over a thousand subscriptions from restaurants, hotels, and other establishments.
•
Insights:
◦
Q: I'm curious how they were able to create such a massive product. It seems like the initial costs would be significant. A: The founder originally worked for a broadcasting/media company. Perhaps due to his profession, he had a keen ability to control costs. Towards the end of his career, he became interested in such solutions. At that time, he established partnerships with wealthy individuals who supported him financially and collaborated closely with him. Thus, the ability to create this product was made possible by (1) having ample time after retirement and (2) having substantial financial support from wealthy individuals. The product you see now is the third-generation version.
◦
Q: It seems like the founder had a strong belief from the beginning that this product would succeed. A: That's correct. They had a fair amount of confidence in the beginning. To create such an innovation, it seems that four elements are necessary: customer needs, feasibility, degree of problem-solving, and profitability. Naturally, they conducted detailed analyses of these elements. (Just as I was about to ask how they conducted such thorough analyses, another person entered the conversation.)
Two things stood out to me:
1.
This company is a true "wine tech" company (la winetech). From a Korean perspective, such an item itself might be considered overengineering. However, they truly drink wine like water. In Korea, unless a place is specifically dedicated to wine, such as "One Glass of Wine Today," it is difficult to find places that offer wines by the glass. Only recently, some restaurants started offering one or two options to be served by the glass among various wines. However, the principle of wine consumption is to finish a bottle within a day or two and preferably within seven days, as the turnover is not high. In France, especially in restaurants, it is common to have at least ten types of wine available. In most cases, there are options of 15cl (150ml), 30cl (300ml), and 50cl (500ml). Due to such demand, there arises a need for a machine like this.
2.
When I first visited this booth, I was busy trying to understand what the product was. But upon further thought, it seemed a bit odd. I, too, have always imagined such items, but the reality is that there have always been practical barriers preventing us from creating such "masterpieces." The way to overcome these practical barriers has always been through a "lean startup" approach. It involves boldly giving up on things you cannot do, breaking down the steps, and minimizing failure while tapping into opportunities. Were these entrepreneurs following this path? So, before leaving the event, I revisited their booth and asked how they managed to accomplish what they did. Sometimes, instead of cautiously progressing, it is better to experiment with how far one can go or how much customers would like it, rather than constantly testing the waters through experimentation and pivoting. Some companies have achieved remarkable success by bravely pushing forward without pivoting. They seemed to be one of those cases.
lululab
•
Process of discovering the item:
◦
The company originally worked with image processing. However, they became interested in the cosmetics industry, thinking that there would be a lot of image data available in that field.
•
Problem to be solved by the product:
◦
The difficulty of selling cosmetics online lies in the fact that it's challenging for people to see how the products look on their own skin. As a result, many people end up purchasing cosmetics based on the inertia of using "what they used before."
•
Solution provided by the product:
◦
The software corrects the phenomenon of facial appearance varying based on ambient lighting and analyzes the skin and skin tone after removing the filters commonly found on smartphones.
◦
When actually using the application, it accurately detects dark circles, pores, blemishes, and skin issues, even in poor surrounding environments.
◦
Cosmetic brands, as well as supply companies like Amway, are considered customers. Their business model revolves around providing software that accurately analyzes facial skin based on images to cosmetic brands in the form of SaaS.
•
Insights:
◦
Q: You mentioned that not only cosmetic brands but also customers themselves show a tendency to stick with their existing products. It seems that the usefulness of the software has not been directly validated through market feedback. Were you not afraid of challenging such a market? A: That is a tricky point. However, even if the item does not receive immediate response, we believe that the ability to collect data and utilize it for sales is sufficient.
Although I didn't have much time to spend with this startup, it seemed that they didn't start with having the technology right from the beginning. When I got home and looked up the company, I found out that they had an annual revenue of 6 billion KRW. I'm curious about how they are making money and excited to see what unexpected forms their success will take.
Studio Lab
•
Problem:
◦
When managing an online shopping mall, it is not easy to take well-posed model photos, and designing the product detail pages to match those shots also requires a lot of effort.
•
Business idea:
1.
When a model uploads images of themselves wearing clothes, the software analyzes the model's pose, atmosphere, and other factors to automatically arrange the photos and design elements for the product detail pages of the online shopping mall.
2.
Additionally, they are developing separate solutions for analyzing the fabric of clothing using special camera-based technology and automatically photographing clothes on mannequins from various angles.
•
Insights:
◦
Q: Are there actual customers using the service? A: We have not yet secured users. However, we are developing projects in collaboration with a large retail company (mentioned the name of a well-known retail company familiar to Koreans).
◦
Q: How were you able to develop the product to this extent and collaborate with major shopping companies without securing users? A: First, we advertised. We wanted to see if we could get users. As you know, it's the methodology that the Toss founders mentioned. It was for the purpose of seeing if users are willing to pay for it. If there is an explosive response, it means it's a successful product. However, offering it for free is meaningless. We needed to confirm whether users would pay for a service that builds product detail pages for them. In the beginning, we manually built each page ourselves.
◦
Q: In performance marketing, you can target specific audiences, but it seems challenging to specifically target online store operators. A: We used Facebook performance marketing tools because Facebook includes Instagram. Since most fashion store operators use Instagram, it was not difficult to reach the target audience.
◦
Q: I'm curious about how the content is arranged. It seems like a rule-based approach rather than a machine learning-based approach. A: We have both rule-based and machine learning-based approaches. Rules mainly handle how to arrange images and fonts. Image analysis is based on machine learning.
◦
Q: I'm curious about how you ended up working in this domain. If there were results from ideation, how did you select the items? A: I don't think entrepreneurs should limit themselves to a specific domain. I think we conducted about 200 ideation sessions, regardless of concerns.
◦
Q: You mentioned that you conducted around 200 ideation sessions. What criteria did you use to filter out ideas initially? A: It seems like it was the prototyping phase I mentioned earlier. Customers are not particularly interested in whether we use AI or not. If there is a service that can build product detail pages for them, customers are willing to pay for it. In the beginning, we manually built each page ourselves.
◦
Q: You mentioned that you were previously at Samsung and then spun off. In fact, have you ever thought about joining a large corporation first in order to seize opportunities, even if it means sacrificing your own desires? A: Especially if the purpose is to seize opportunities, I would resist joining a large corporation. There are tens of thousands of Samsung employees, but there are fewer than a thousand people in venture companies, and there are not even 30 teams formed in a year. Q: Despite that, why do some people who want to start their own business still prefer to join a large corporation first? A: I think it's because they want to have a startup story. Also, I think it's because they need to understand the customers. For example, if they want to target the market of semiconductor supplies, they should go to places like Samsung Electronics, which is the best at semiconductors in South Korea, to see such things.
The two most impressive points can be summarized as follows:
1.
To confirm the demand for such automation projects, they advertised and expected the results to be at the level of AI-generated results that could be practically implemented. I imagined designers diligently completing the product detail page design work one by one, even for projects that seemed quite sophisticated and impressive. Even for such projects that may appear remarkable and impressive, it was evident that they had to type and walk through each step without significant sponsors.
2.
Actually, without properly seeing the result of the product, I had the thought, "Is this really necessary?" However, when I saw the result, I could see automatically generated product detail pages with designs that would easily fit in a trendy fashion magazine. I was amazed by the quality of the result.
Manta
•
Product:
◦
The company develops a personal mobility device that can be used on water, utilizing a motor, battery, and pedals. It can travel on water at a speed of approximately 22 kilometers per hour.
•
Insight:
◦
Q: Is this company's product the only one they have? A: Yes, that's correct. They have no other products besides this one.
◦
Q: How is it possible to create such a masterpiece? It seems incredibly difficult to create such a work without any capital, so I am gathering stories from entrepreneurs who have created such works. A: The founder of this company is a person from New Zealand who was very wealthy. They sold their previous business for a high price and received support from their surroundings, using those funds to realize their lifelong dream.
Similarly, I realized that without market validation and with the goal of creating a masterpiece, it requires either a significant amount of money or a strong conviction from one's background. In the case of this company, they had a substantial amount of money.
Arcade for Good
•
Product:
◦
It is a game console that requires a small voluntary donation to play arcade games. The console is equipped with payment hardware, and donations can be made easily through NFC (in Europe, most card payments are based on NFC).
•
Insight:
◦
Q: Did you create this from start to finish? A: We have a background in development. (Showing the computer inside the machine) We created the software that runs on this small computer. However, we didn't create the shell of the game console.
◦
Q: How many units have actually been distributed, and how much do people play the games on the distributed consoles? A: So far, we have distributed about 30 units. We placed them in places like gas stations, airports, and subways. People play games on them around 8 times per day.
◦
Q: What is the biggest bottleneck in pursuing this business? A: The price of purchasing the devices is the biggest issue. It costs about 2,000 euros to make one device. However, if we can produce multiple devices at once, the price decreases significantly. So, we are trying to secure funds to purchase a large quantity at once.
◦
Q: I'm curious about the business model. If someone donates around 1 euro, what percentage actually goes towards donations? A: Currently, 100% of the donation goes towards charity. In return, we sell the machine itself and have chosen to insert advertisements during gameplay.
◦
Q: I'm curious about the reason for starting this business in the first place. A: I believe the gaming market in France is quite large. There are still quite a few people who enjoy playing these classic video games. I don't know exactly why we chose this type of machine, but wouldn't you want to go and try it out if such a device existed? I think that's the reason.
I was impressed by the fact that if they have about 8 games played per day, the game usage is quite substantial. At this level, it far surpasses the average usage of a shared electric scooter in South Korea. If the average payment amount per person is 1 euro (approximately 1,400 won), and around 20% goes towards operational costs (average for charities), it results in a daily revenue of 2,400 won per device. If the cost of producing the machine decreases to 500,000 won per unit due to mass production, the initial investment can be recovered in about 200 days. I'm not sure about the criteria for social enterprises, but if this model is viable in South Korea, it could provide a stable cash flow for a small team. With 1,000 units distributed, it would generate 60 million won per month or 700 million won per year.
Meta
•
Product:
◦
A product that implements a virtual workspace in VR space, along with the Oculus PRO hardware used to realize it.
Apple recently released VR devices. Apple is always an expert in turning "young technology" into a product that is close to perfection, maximizing its value as a commercial product. This was the case with laptops, smartphones, and wireless earphones. In other words, I perceive Apple as a pioneer in the commercialization of technology. The Oculus Quest, a cost-effective model from Oculus, is priced around 300,000 won. However, Apple's devices cost over 4.5 million won. As with wireless earphones, Apple devices have often been criticized for being expensive, but Apple has always achieved the highest sales figures. To experience a device that is on a similar level to Apple's, I tried the premium model Oculus Pro from Meta, priced at 2 million won instead of the cost-effective model. It was an attraction where I had to wait for more than 1 hour and 30 minutes after making a reservation. I experienced various activities with a few other Meta employees in a virtual conference room. The level of completion was quite high, which was impressive. The spatial sound effects based on character positions, hand gesture recognition technology, smooth frame rate, depth perception, and stability without motion sickness were all beyond my expectations. Currently, public opinion mainly revolves around the notion of "Who would buy such an expensive VR device." However, I was convinced that the completeness of Apple's VR device would be higher than what people think.
Pixellot
It was a conversation with an Business development VP Brian Phillpotts.
•
Product:
◦
Develops a system for creating sports broadcast screens without the need for cameramen. Cameras automatically capture the game from suitable angles, and the system handles both media mode and broadcasting mode. It also includes capturing and analyzing team plays for the purpose of analysis and improvement, known as strategy mode.
There were 3-4 key points that left an impression on me and sparked my thoughts:
1.
Although I can't recall the exact details, the product was already widely used. When I talk to people who have extensive experience and are involved in large-scale businesses with high levels of completion, I often feel overwhelmed. I felt a similar sense of awe during my conversation with this company.
2.
The idea of manipulating camera angles to fit more advertisements into the screen and increase revenue beyond the current level was something I never imagined.
3.
Some business domains have high barriers to entry. For example, goals like "creating a third smartphone company after Apple and Samsung" or "developing a remedy for eternal life" are difficult to achieve as my primary objectives. I refer to these as things I "cannot currently do." For instance, if I wanted to automate sports broadcasting or recording, I would need to have direct access to cameras at sports venues. However, even with lower-tier or amateur games, it's not easy to have the opportunity to handle cameras and conduct various experiments. Thus, companies developing products for such domains likely started with solutions in domains with lower hurdles that they could easily access.
4.
If I were to enter a domain with high barriers to entry directly, what would I do? Perhaps I would learn the basics of filming through school broadcasting clubs and use 3-4 smartphones as makeshift cameras to willingly film and edit amateur games. I would provide the footage for free to the organizing bodies of the games, monitor the reactions, accumulate know-how, and build references to establish a solid foundation for gaining trust on a larger stage. Only then would I consider approaching the larger market for potential partnerships.
Equip
•
Product:
◦
Produces and sells unmanned rental boxes for people who visit parks without the necessary sports equipment.
◦
Distributes the product to locations such as hotels and parks, using Bluetooth-based unlocking for the rental boxes.
◦
Utilizes a system similar to app-based vehicle rental services like SoCar or GreenCar, where the next user confirms the proper usage by the previous user.
◦
The company sells only the rental boxes and the mobile platform. It's the responsibility of the buyer to purchase the items to be placed in the boxes and replenish any lost items.
•
Business Model:
◦
Sells stations (rental boxes).
◦
There are two types of stations:
▪
Type 1 stations do not charge installation fees but allow the buyer to take 50% of the rental revenue, while the company takes 30%.
▪
Type 2 stations require a €1,500 installation fee, but the company does not take a share of the rental fee.
•
Journey to Finding the Product:
◦
The company was initially focused on snowboard rental business as an internal venture. However, they faced challenges such as frequently broken plates. As a result, they decided to pivot to renting out other sports equipment instead of snowboards.
•
Insights:
◦
Q. What are the most challenging aspects in operations and development? A. Software issues seem to be the most challenging.
◦
Q. Were there any legal issues with the installation of the facilities? A. Since the government or hotels purchase and install our product, there should be no legal issues.
◦
Q. Is it battery-operated? I don't see a power plug. A. Yes, it is. It can be used for two years without recharging. We initially used solar panels, but it was not cost-effective, so we opted for rechargeable batteries.
◦
Q. It seems like you would have followers. Are there any similar teams working on similar items? A. There is one team in France and one in Germany. However, they mainly focus on fitness items in their businesses.
◦
Q. Is there any plan for expansion into Asia, where such items are also needed? A. We would like to, but we want to establish a stronger foundation in Europe first. We are a Swiss startup, and I come from Germany. The company seems to be focusing on expanding in the German market.
◦
Q. What did the respondent study? A. I studied sports business. Q. Does that help with your current work? A. (Explains various aspects) I'm not sure if it's particularly helpful, haha.
It seems like a good idea to give it a try as one of the various items in Korea. It doesn't appear to be something difficult or complex. It's also not hard to imagine how to pivot into different rental items.
i-Virtual
•
Product:
◦
The product is capable of extracting various health information when taking a selfie that includes the shoulder area. It can capture data such as heart rate, stress level, and respiratory rate.
•
Insights:
◦
Q. Where can such a solution be sold? A. It can be sold to telemedicine platforms or insurance companies.
◦
Q. What was the most challenging aspect? It seems particularly difficult to enter the early market, such as the healthcare market. In Korea, telemedicine is still regulated by law. A. Gaining trust is the most challenging aspect. However, entering the early market was not as difficult as expected. France was similarly challenging. Just a year or two ago, there was no such thing as a telemedicine market.
◦
Q. The technology is quite impressive. I'm curious about the team that started developing this item. A. The initial team consisted of two engineers, one researcher, and the CEO.
I couldn't help but be amazed by the technology itself unless it is a lie. It felt similar to identifying people in satellite photos. Even if there is a lot of information on a person's face, how much of it can be transmitted through a smartphone camera, considering the subtle tremors caused by breathing and heartbeats?
MOPSO
•
Problem:
◦
In this world, money laundering and illicit financial activities occur frequently. This product plays a role in detecting such illicit money.
•
Insights:
◦
Q. Frankly, I have no knowledge in the financial industry, but I'm curious about how the product detects money laundering and whether it's feasible. Could you provide some background about yourself as the CEO? A. Detecting money laundering is not an easy task. Even if we ask people to provide information to detect money laundering, they may not willingly provide all the necessary information. That's why I often introduce myself as a philosopher with a strong interest in psychology. I have worked in the banking and financial domain for 10 years.
◦
Q. Based on the displayed screen, it seems like you represent the flow of money in a graph format. If a single node in the middle is disrupted, wouldn't it be impossible to detect money laundering? A. (He joked, asking if I wanted to learn how to launder money. Then he mentioned that there are many ways to launder money, but one of the best ways is through a restaurant because it provides an easy way to insert legitimate payments into illegitimate transactions) Yes, those aspects are indeed challenging, and we are a startup that aims to detect such activities.
Considering the nature of the financial industry, it seems that market entry was extremely challenging. Building trust with the financial sector, which tends to be resistant to accepting new things, was the most difficult aspect. This led to my curiosity about how they conducted the Proof of Concept (PoC). Naturally, I suggested that they first demonstrate the system's functionality by showcasing successful operations using demo data. On the other hand, I also considered that the founder's 10 years of experience in the financial industry and their specialization and connections might have played a role in enabling them to take on such a challenge.s
UPPLY
•
Problem:
◦
Logistics transportation costs are constantly changing depending on the type of goods, mode of transportation, and timing.
•
Item:
◦
Analyze prices to predict future logistics cost trends.
◦
Provide a clean dashboard to support decision-making in logistics operations.
•
Insights:
◦
Q. How did you obtain real-time logistics transportation cost data? I'm also curious about the background of the CEO. A. The company itself started from a large logistics company called Geodis. The CEO had a background in consulting, and they began developing the item using data from Geodis.
I can empathize with the importance of strategies for moving goods at the right time and using the appropriate mode of transportation, considering that transportation costs are not fixed. However, it seems that there are already established companies that can create such platforms. I sought out this company to learn about ways to penetrate markets that are not easily visible to consumers, such as the logistics market, but I didn't gain any significant insights.
BORDERLESS
•
Problem:
◦
Many Korean fashion companies want to create a foreign atmosphere or the atmosphere of overseas brands.
◦
Typically, fashion brands have internal marketing teams that go abroad for overseas shoots, incurring significant costs. They may also bring foreign actors/models to Korea, which also comes with high costs.
◦
Even for highly famous models, the model fee significantly decreases when (1) the shoot takes place near the model's location instead of bringing them to Korea and (2) when relatively unknown Korean brands hire them.
•
Process of Identifying the Problem:
◦
The CEO originally worked in the plant business overseas. He witnessed Brazilian models going abroad for shoots and thought, "Can't we just bring the clothes there?" This led him to start contacting and meeting with various brands.
•
Insights:
◦
Q. From what I heard, models don't seem to be tied to specific companies through contracts. A. That's correct. Models are not necessarily exclusive to specific brands. However, there are agencies that manage models.
◦
Q. In the past, didn't the marketing teams of Korean fashion companies go abroad for shoots to precisely capture the desired atmosphere? If we simply send the clothes abroad for shooting, wouldn't it be challenging to control those aspects? A. That's correct. Initially, we only sent clothes and entrusted everything to local productions. The video you are seeing now is from the early days of the business. It doesn't give the same feeling preferred by the Korean market but rather a vibe that would resonate in the U.S. market. We received feedback from clients that the mood didn't align with what Korean customers liked. So, we started getting involved in the planning process. This seems to be our know-how.
◦
Q. What does a creative director do? Could you also share your background? A. As mentioned earlier, I am involved in the planning process. I thoroughly analyze everything, including the direction of light, weather, model selection, clothing, background, lighting considerations, the overall atmosphere considering client requirements, references, and pre-shoot surveys. This ensures that even if the shoot takes place with a local production, the results align with what Korean clients want. This is not where I started my career; I have worked in the fashion industry for a long time.
◦
Q. What do you think the next scale-up would be? A. Currently, it is quite distant from full development. However, we have a team member who has some knowledge of development, and they are well-organized with data on shooting methods, requirements, lighting, etc. Now, we think it's time to develop it into a B2B platform, but it's still a bit far off. Additionally, if we previously assisted Korean companies in going abroad, we plan to expand to foreign companies shooting in Korea.
Inspeak
•
Problem Being Addressed:
◦
When developing new drugs or other substances, various liquids need to be mixed, and their reactions observed. However, these experiments can be time-consuming, as completing a full cycle can take a significant amount of time. This means that if something goes wrong in the middle of the experiment, one must wait until the entire process is completed.
•
Product:
◦
An extremely affordable liquid analysis chip. The chip analyzes the liquids in real-time as the reactions progress.
◦
The chip is small and inexpensive, allowing for the construction of large-scale experimental environments and running multiple experiments in parallel.
•
Insights:
◦
Q. It seems like the development costs would have been significant. I'm curious about how you started in the very early stages of the startup. A. The founder originally pursued a Ph.D. in semiconductor technology at MIT in the United States. During his studies, he came up with this idea, and after returning to France and studying further for a few years, he presented the idea to investors and secured funding.
◦
Q. Did all the team members have backgrounds in chemistry or semiconductors? A. The team members all came from different academic backgrounds, pursuing their Ph.D. or Master's studies in various fields. Some had a physics background, while others specialized in optics and engineering.
This case demonstrates experts coming together to solve their own problem, resulting in the development of a product related to laboratory experiments.
Vinidaily
(Co-)Founder이자 CMO Carla Carfagno 와의 이야기였다.
•
Problem Being Addressed:
◦
In Europe, there are numerous wine options available, and individuals have diverse preferences. However, it's impossible to try out every wine and explore the wide range of options.
•
Product:
◦
The product responds to a few specific questions, such as preferences for sweetness, fruitiness, or strong tannins in wine.
◦
By taking a picture of the entire wine shelf using a smartphone, the screen displays recommendations for specific wines.
•
Insights:
◦
Q. I'm curious about the team's background. A. There were four co-founders. I graduated from a finance school, while the other two co-founders had business school backgrounds. The remaining team member came from an engineering school.
◦
Q. I think wine recommendation apps are quite common. Personally, I use the wine application Vivino a lot. What do you think is the biggest difference between your product and Vivino? A. Vivino requires scanning each wine individually. However, as mentioned earlier, we can take a picture of the entire wine shelf and identify all the wines within it. This is the technology we have developed.
◦
Q. I assume that a lot of research and wine label image data acquisition was necessary to develop the product. How did you have the confidence to proceed, believing that the product would succeed? A. Two years ago, we created a chatbot messenger that could only be used in a wine shop selling about 100 different wines. We monitored how users interacted with the chatbot and their satisfaction with the results.
◦
Q. How did you acquire the wine label image data? A. Initially, we collected the data ourselves. (They may have later outsourced the data collection to external firms since there are many data outsourcing companies available.)
◦
Q. Even if you have the wine label image data, how did you acquire the data about the taste profiles of each wine? A. There are various organizations, such as the International Organisation of Vine and Wine, that have wine-related data. We were able to gather data from these organizations.
◦
Q. I'm curious about the business model. A. We plan to use a subscription-based model. Currently, the service is free.
Two points stood out to me:
1.
The decision to limit the scope of the application to a single wine shop (an ODD) and use a chatbot as a tool to assess market response to the product was impressive. It allowed them to collect data on various types of wines without the need for camera-based identification, as they could simply transfer all types of wine data to an Excel sheet from a single wine shop.
2.
As mentioned before, the high frequency of wine consumption, with one person consuming one or two bottles of wine each week, made me realize that this business model could also work as a subscription-based service.
</aside>
It can be overwhelming to be surrounded by such a vast selection of wines in any random French supermarket.
Unboared
People are enjoying gaming using their smartphones as joysticks while looking at a shared monitor.
I had a conversation with Victorien Defosse, the co-founder and CEO.
•
Product
◦
Play casual games using mobile devices as joysticks while looking at a shared monitor.
◦
Both the mobile devices and games are web-based.
•
Insights
◦
Q. It seems like the idea and the game itself are quite simple, but there were probably several feasibility testing stages before bringing them to life. I'd like to hear the initial story. A. Initially, we created the simplest platform. Although there were many sophisticated ideas, we pondered over what the main feature of our core product should be. It turned out that the ability to play using mobile devices as joysticks was the most crucial feature. To test this idea, we created a prototype. FAKING THE PRODUCT! We created a fake game platform, a fake mobile joystick platform that seemed connected but wasn't actually connected to others. It took just two weeks to complete the MVP. We didn't have a clear target audience in the beginning since this game platform could be used by adults, children, and companies alike. We created around 10 focused hypothesis groups and continuously tested which target group felt the most appealing. My advice would be to build the product within two weeks and test it as quickly as possible. Make the product as cheaply as possible, even without writing code, create test groups, and experiment with them.
◦
Q. How did you measure whether people were willing to pay for and enjoy the product? A. We are currently in the testing phase. We knew there was demand, but we are testing which monetization method people are more willing to pay for.
◦
Q. It seems that people need to gather in one place and have a shared monitor to enjoy the game. Marketing and advertising must have been challenging in that regard. A. It may indeed be challenging to market it solely as an app. Also, customers who encounter advertisements may not have an immediate access to a monitor. However, by using advertising platforms like Instagram, we can show simple demo videos and attract potential customers who may be interested in using the product later.
The software allows multiple people to play together. The idea seems interesting, and I expected there to be something special in the validation process since the item requires multiple participants to experience synergy. However, the emphasized points were the same. The CEO mentioned that "it is really important and I wish someone had guided me in these things when I started. If there were someone who could provide guidance, it would have been great. Unfortunately, there were no such people for me. If you have any questions, feel free to reach out on LinkedIn," while handing me his business card. I was truly grateful. Furthermore, I found it impressive that even for a product that seems difficult to test, they followed the recommended practices of lean startups. They might even advise deploying every two weeks.
QuantumCat
•
Problem-solving product
◦
They are a catalysis company.
•
Team
◦
The project originated from research conducted during a PhD program at KAIST (Korea Advanced Institute of Science and Technology). They received investment from Samsung.
•
Insights
◦
Q. Since I'm in the software domain, it's hard for me to imagine where such a product is developed. It would be helpful if you could provide more insight to improve my understanding. A. You can think of a chemistry laboratory where we conducted experiments in middle and high school. There are various raw materials and equipment for analyzing them.
◦
Q. In the software domain, it is possible to create a product and receive feedback, but for hardware or chemical products, it seems harder to have the mindset of "Let's try it because there is enough potential!" I'm curious about the decision-making process you go through before taking the plunge. A. To be honest, it's a leap of faith. The software domain has the advantage of having a faster feedback cycle, but it's a bit more challenging for hardware or chemical products. However, that doesn't mean we dive in without any justification. We develop the product with the mindset of addressing a specific problem or point of improvement. However, that alone is not enough. It seems that success also depends on the alignment with social trends and the government's investment direction. It's practically impossible to accurately predict all of these factors. If you want to be completely sure about them, I don't think you should start a startup. That's why I mentioned it's a leap of faith.
(Note: The term "모 아니면 도" in Korean can be translated as "either everything or nothing," indicating that there's a level of uncertainty and risk involved in pursuing such endeavors.)
ArtCentrica
I had a conversation with CEO Marco Cappellini.
•
Product
◦
An online platform specialized in art education. It includes a wide range of features.
•
Insights
◦
Q. The product seems to have a variety of features. What criteria were used to add these functions? A. Some of the features were derived from my personal experience, while others were collected based on the feedback and requests from teachers during our interactions.
◦
Q. I'm curious about the business model. A. It is a B2G SaaS (Business-to-Government Software-as-a-Service) model.
◦
Q. It seems like it would have required a significant investment. I'm curious about the initial story behind it. A. Ah, we have a parent company called 'Centrica' that deals with education as a whole, although we specifically focus on art-related aspects. Therefore, we didn't develop the product in a state of scarcity.
While anyone can create features such as comparing two images side by side or organizing a timeline, what amazed me was the ability to extract specific decorations, patterns, and details from the artwork and compare them. It also allows for a close-up view of the brushstrokes, providing a detailed digitalization of the artwork. I wonder how deeply art is taught in Italy to gain market recognition and sales for such features.
CLIKA
I had a conversation with CDO Alessandro Mapelli.
•
Problem
◦
Existing solutions for optimizing deep learning models, such as TensorRT and PyTorch quantization tools, are slow and not stable for all models.
◦
Companies with their own data and models do not want to rely on SaaS solutions provided remotely through web APIs.
•
Product to Solve the Problem
◦
Developing a Python package that operates on-premises.
◦
Using a slightly different quantization method than commonly used methods for model optimization.
•
Process Leading to Identifying the Problem
◦
The CTO is from Israel and previously worked at Mobileye.
◦
While working at Mobileye, they were able to identify optimization issues.
•
Insights
◦
Q. I'm curious about the background that led to the development of this solution. A. The CTO is from Israel and has experience at Mobileye. That's how they were able to identify these issues. We built everything from scratch without using any open-source frameworks, so it works slightly differently from other quantization tools.
◦
Q. I'm curious about the revenue model. A. We charge based on licenses that can be used on one device. It costs $100,000 per year. It may seem expensive, but for companies that constantly need to run models, the difference in infrastructure costs between running optimized models on CPUs and GPUs is significant.
◦
Q. Looking at the performance table, it doesn't seem like there is a significant performance difference compared to TensorRT or PyTorch-based quantized models. Why is this level of performance improvement necessary? A. You're right. Each case has different levels of success with different methods. Even if a method seems to work well, it's difficult to know if it really does. Our solution performs more stably than TensorRT or PyTorch-based quantization for any model. Additionally, the time required to generate optimized models with TensorRT is very long. We can perform a task that takes 30 minutes in just 7 minutes. We want to enable any junior AI model developer to use the model without worrying about optimization.
◦
Q. Are there any companies currently using this solution? A. Not yet, but we have completed a closed beta test with a relatively large company in Korea that wanted to run the RetinaFace model. These companies don't want their models or data to be exposed externally. That's why we created an on-premises solution instead of a cloud-based one.
◦
Q. (Considering that optimizing models offline and using the optimized model file wouldn't be a significant issue in terms of time) Can users receive only the model files? A. It's not possible to receive only the model files. Q. So, each time someone wants to use the model, they have to go through the optimization process at that time? A. Currently, yes. I believe this feature will be added in the future.
◦
Q. It was mentioned that the representative company is in Korea. Is it a Korean company? A. Yes, it is. The headquarters are in Korea. Q. Then, are most of the employees working in Korea Koreans? A. We have 11 employees, 5 Koreans, and 6 foreigners. We communicate in English within the company.
◦
Q. I'm curious why the CDO is in Korea. Do you think Korea is better than Italy in terms of the startup ecosystem? A. The CEO is Korean. I lived in Korea for five years. I think Korea is much better. Hahaha. I believe Korea has a much better startup ecosystem. The government is making a lot of efforts to provide funding, and there are accelerator programs, such as the one with Samsung, that are active. I think countries like France and Switzerland have excellent startup support. However, when it comes to the initial capital-intensive development, I feel that the support structure for startups in Italy is weak. I also think that the tax structure in Italy is quite challenging.
◦
Q. You mentioned that you are a CDO. I'm curious if there are any design elements involved in the model optimization process. A. We are developing a platform that allows monitoring of deployed models. Design elements are required for this platform. Of course, it is included in the license price.
Proto
•
Product
◦
The company creates fascinating 3D monitors.
◦
A regular 2D monitor is attached to the top of a triangular box. People simply need to look at the triangular box. The light emitted from the 2D monitor creates a highly immersive effect, giving the impression of a 360-degree panoramic view.
•
Team Background
◦
The team consists of individuals with backgrounds in optics and product design, working on product development in Silicon Valley.
•
Additional Insights
◦
Q. It's really amazing. However, I'm curious how you can create a sense of depth with just a 2D display. A. You're right; we have installed a 2D display. We analyze the emitted light and modify it to make it appear three-dimensional to our eyes.
◦
Q. What market/application are you targeting with this product? A. Initially, it could be sold to the defense industry. It could also be used by stock traders who want to have a single large screen without the need to switch between multiple monitors. It can also be used in vehicle displays or gaming, as demonstrated here.
◦
Q. I believe VR devices, with recent releases from companies like Apple, will have significant advancements in the future. Don't you consider them to be major competitors? A. We don't think so because we believe they serve different purposes. First of all, wearing something on your head is never a comfortable form.
When observing this booth, I had two main thoughts:
1.
I felt that this approach is a complete technology-driven startup approach where they create something incredible first and then find customers who need this technology and product. I think they must be aware that as long as they are not fools, they can venture in this way. However, recently, both globally and in Korea, "lean startup" has been considered the answer for ventures. I'm curious about how these technology-driven startups survive.
2.
From the perspective of experiencing this product and Facebook's VR devices, I found Facebook's VR devices to be much more comfortable. The field of view was too narrow in this product. To fully enjoy the display, you had to stand in the right position, similar to holographic technology. Even moving slightly further away or looking in a different direction by about 15 degrees would diminish the display performance. It made me think that VR could be more suitable for areas such as military or unlimited-monitor computing, where the limitations of this product could be overcome.
</aside>
Tidal
•
Product
◦
A combination of underwater hardware and software that can analyze fish in fish farms.
•
Problem to Solve
◦
In fish farming, it is important for fish to grow evenly. If some fish start consuming slightly more food, it can disrupt the balanced growth of all the fish.
◦
Fish farmers currently lack a way to verify whether the fish are growing evenly (i.e., if they are receiving the correct amount of food).
•
Additional Insights
◦
Q. Looking at the screen, it seems like there are not just one or two fish. (Based on the tagging of ID values on each fish) It appears that you can distinguish each fish. Is my understanding correct? A. Yes, that's correct. We have put in a lot of effort to distinguish each fish. Through extensive research, we discovered that each fish has unique patterns on its side, which allowed us to differentiate them. We invested a significant amount of resources into this aspect, but unfortunately, the clients were not interested in such details. They were more concerned about the overall distribution. So, our focus shifted to inferring the weight of the fish displayed on the 2D image and extracting the distribution based on that. Additionally, we can also capture the food that falls into the water when it is sprayed.
◦
Q. How did you consider building such a massive-scale product? A. Fortunately, the largest fish farming company in Norway, Mowi, approached Google to seek a solution. That's how the project began.
mowi
The key takeaways from this conversation can be summarized as follows:
1.
Even large companies like Google can make the mistake of overengineering, as evidenced by this project.
2.
It serves as another great example of using AI technology to solve real-world problems. Particularly, it addresses a massive-scale problem, which makes it even more impressive.
3.
The development of such a massive product was made possible because of the resources provided by Google and the collaboration with Mowi, a large client. It may not be something that I can achieve right now.
4.
On the other hand, a common and persistent challenge for such solutions is the pricing or pricing strategy that comes with making the solution "sufficiently usable and perfect." If it is a solution developed by Google, the price is likely to be very high, and it may not be tailored to small-scale fish farmers. While Korea and Japan have thriving fishing industries (I'm not sure about fish farming specifically), most of the players are small-scale fishermen. If the task is as simple as measuring weight based on a 2D image and if it is an important aspect for small-scale fishermen, it might be worth considering entering the Korean and Japanese markets with a pricing strategy that can appeal to them. Collaborating with organizations like fisheries cooperatives could also be helpful.
Audemic
CEO Joshua Mitcham actively approached me as I was reading the product description.
•
Product
◦
An application that takes a research paper as input and provides a summarized version.
•
Additional Insights
◦
Q. I'm curious about the revenue model. A. It operates on a monthly/annual subscription basis, and the first month is free. Most of our paying customers are individuals studying in master's or doctoral degree programs. Many people are opting for annual subscriptions.
◦
Q. What are the differences between the demo you showed me and the initial PoC version that implemented the early idea? A. While many aspects have been refined since the initial version, the overall concept remains similar. We were fortunate to receive a positive response right away, which is an uncommon case.
◦
Q. I'm curious about the next stage of scaling up. A. We are currently developing the application. We're not sure exactly who our target audience will be, but here's a naive idea. AI research papers, as well as various scientific discoveries, are exchanged in the format of research papers. However, from the perspective of the general public, not everyone wants to understand all of that, nor do they have the time for it. Our aim is to extract scientific discoveries and knowledge from the format of research papers and reconstruct them into a new format.
There was a global frenzy surrounding ChatGPT. People were starting to think about how to commercialize it. In the case of applications that process text-based information, it occurred to me that similar items could be quickly prototyped without writing a single line of code.
Tipsi
It was a conversation with CTO Maxime GIL.
•
Problem
◦
The use of card payments in restaurants and bars is increasing.
◦
When paying with a card and wanting to leave a tip, the tip amount is added to the menu price, resulting in the entire amount being treated as revenue for the establishment. As a result, customers often do not leave tips, even for excellent service.
•
Product to Solve the Problem
◦
Software for card terminals that allows customers to specify a particular waiter or waitress to receive the tip.
◦
The tip added through the terminal goes directly to the designated waiter's wallet, rather than being treated as revenue for the entire restaurant.
•
Additional Insights
◦
Q. I'm curious about what exactly you have built. Did you create the card terminal as well, or just the software? A. We didn't create the card terminal itself. As you know, it's not always a good idea to handle hardware directly. We only developed the software.
◦
Q. If it's software, does it need to be installed on the card terminal hardware? How did you convince the terminal hardware company to cooperate? A. It's simple. We just pay them.
◦
Q. The card terminal on the table looks a bit different from the ones I usually see. The ones I'm familiar with have a screen similar to what I see on the TV. They seem to have different operating systems. Do you have to develop separate software for each operating system? A. Yes, we developed separate software for each operating system. The one on the table runs on the Android OS. The terminal you see on the TV is currently dominating the market. We had to develop software specific to that operating system.
◦
Q. Earlier, you mentioned that you pay the card terminal company. How do they generate revenue? A. For card terminals that use the Android OS, they are somewhat open to this. They charge a fee of 1,500 euros (about 2 million KRW) for registering in the application store, and there are no further charges. However, for installing the software on the currently monopolized terminals, they charge per device.
◦
Q. If a restaurant using the old-fashioned terminal on the TV screen wants to use your tip software, how do you install it? It doesn't seem to have features like an application store. A. They just give us a call, and we go and install it.
◦
Q. I'm curious about how you thought this product would be useful. What testing methods did you use? A. Like everyone else, we ran beta services in a few restaurants. We received positive feedback there, so we are now developing it further.
◦
Q. I'm curious about what kind of restaurants would be interested in this. A. Restaurants where the staff want to work harder and create a pleasant atmosphere. It also helps in recruiting people who want to work at a restaurant.
I'm not sure if it's possible in South Korea to install third-party software on card terminals. It seems that new players are constantly entering the card terminal market in Europe. This startup first developed a relatively heavy task of providing tip software and then entered the market. Fortunately, there were restaurants using Android OS-based terminals, so the development process may not have been too slow, and the product received a good response from consumers. However, if there were no Android OS-based terminals available and consumers didn't like the product, it could have been a project that consumed a significant amount of resources and eventually ended.
oraigo
It was a conversation with CTO Gianluca Maguolo.
•
Problem
◦
Truck drivers are often unaware when they are driving drowsy.
•
Product to Solve the Problem
◦
A headband (EEG) that diagnoses drowsiness based on brainwaves.
•
Process of Identifying the Problem
◦
The CEO mentioned that he had been looking for a device that could prevent drowsy driving since he was young.
◦
However, such a device did not exist, so he wanted to solve the problem himself.
◦
Initially, he planned to create a wristband that could measure body temperature and heart rate to determine drowsiness. Nowadays, devices like Apple Watch or Mi Band are capable of accurately measuring body temperature and heart rate, but these indicators do not immediately respond to drowsiness. It takes a sufficient amount of time after drowsiness begins for changes in body temperature and heart rate to occur.
•
Additional Insights
◦
Q. It's an impressive product. Have you already sold it to customers? A. No, it's still in the prototype stage, and we haven't launched it yet. However, we have received investments. We believe there are enough sellers who would be interested in selling such products. (They seemed confident and not worried about it.)
◦
Q. What market are you targeting? A. Currently, we are in the stage of finding product-market fit. Initially, we plan to sell to trucking companies and hospitals, and the B2C market will be our final entry point.
◦
Q. I'm curious about the people involved in developing this product. A. Since we have to deal with both software and hardware, the team members come from diverse backgrounds. We have a friend who majored in both computer science and neuroscience. They introduced us to EEG to the other team members. As for hardware production, we are outsourcing it to the United States. It's quite expensive.
◦
Q. You mentioned that after reading the brain signals, you plan to provide an API for developers to utilize. I didn't realize that BCI (Brain-Computer Interface) technology has advanced to that extent. I wonder if Elon Musk is considered your competitor. A. Unfortunately, Elon Musk didn't visit our booth yesterday, haha. The difference between our approach and theirs is that they mainly focus on receiving signals directly from the brain. It may sound a bit brutal, but they would be able to read much clearer signals than us, who read signals from outside the brain. Although they could be considered competitors, the sectors in which the technology is applied would be different, so I don't particularly see them as competitors.
The problem clearly exists, but I'm not entirely convinced of the market demand or the solution's ability to effectively address the issue of drowsy driving among truck drivers. I'm curious about how the company will grow and how they will approach the market.
글을 쓰는 데 참고한 자료입니다.
1.
작성 중입니다.
글을 쓰는 데 반영된 생각들입니다.
1.
작성 중입니다.
이 글은 다음 글로 이어집니다.
1.
작성 중입니다.