EQUNAUTS Interview #01 : Sadahiro Yoshikawa, Research Engineer
Equmenopolis has a mission to “deploy conversational AI agents in various educational and work settings to enhance creativity and productivity.”
We provide “LANGX®︎”, a service that assesses speaking skills through English conversations with a conversational AI agent. Additionally, we develop the AI platform that serves as the technological foundation for this service.
At Equmenopolis, a diverse group of members with various expertise and backgrounds — known as “EQUNAUTS” (explorers of future societies) — come together, leveraging their skills and experiences to make an impact.
This time, we spoke with Mr. Yoshikawa, a research engineer, about his current work and the unique aspects that make Equmenopolis an exciting place to be.
For more details about “LANGX®︎”, click here ↓
— First, could you tell us about your work at Equmenopolis?
I work as a research engineer, focusing on the research and development of conversational AI.
On the research side, I participate in academic conferences and write research papers. On the development side, I manage the entire dialogue system development process, enhance and improve dialogue functions and infrastructure, and promote the systematization of dialogue quality management.
— What led you to join Equmenopolis?
Originally, I was a regular software engineer with no experience in building conversational AI.
At one point, I wanted to create a service that could alleviate anxiety through real-time conversations with a system. To pursue this, I transitioned from a full-time employee to a freelancer, reducing my workload so I could use my free time to develop a prototype of a service that would listen empathetically to users about their day.
However, after building the prototype, I realized I didn’t fully understand what real-time dialogue should be like to achieve my goal. I debated between studying Psychology and Computer Science and ultimately chose computer science. At the time, ChatGPT hadn’t been released yet, and conversational AI was still underdeveloped. I also recognized that machine learning was essential for building advanced systems.
To further my research, I enrolled in a computer science graduate program that accepted working professionals while continuing my freelance engineering work. My professor at the graduate school happened to know Dr. Matsuyama (CEO), and through that connection, I joined Equmenopolis as a researcher and engineer specializing in conversational AI.
“I had never seen a conversational system as sophisticated as this before.”
— What aspects of Equmenopolis caught your interest and made you decide to join?
I had previously seen a demo video of InteLLA’s dialogue system, and I was impressed because I had never seen such a sophisticated conversational system before. Before graduating from graduate school, though I was still inexperienced as a researcher, I wanted to join a company that was developing such systems. So, I decided to take a leap and apply.
“It allows us to tackle areas that are still unexplored in research.”
— As a research engineer, what is the most exciting moment for you, and what is the most challenging moment?
It’s challenging because not only are engineering skills and knowledge required, but also various other areas of expertise, such as machine learning. However, that’s also what makes it exciting.
When a system developed in research is finally used by real users, solving issues that arise depending on the user and the context — such as where and how the system is used — can be difficult. Continuously improving the service with actual users is fun, but it is definitely challenging.
For example, we provide a service that assesses English speaking abilities, but the conversation patterns between advanced and beginner English speakers are completely different. Additionally, real-time conversations are not just about responding quickly — they require ensuring network communication quality, detecting noise levels, performing accurate speech recognition, and adjusting conversation tempo to effectively engage with the user. Compared to typical web services, these factors make it much more complex.
Our service demands the creation of a robust conversational AI capable of making accurate assessments in such situations, which is highly challenging. However, this area of AI development and quality management is still relatively unexplored in research, and that’s what makes it very interesting.
“I want to make conversations with AI more positive.”
— Could you tell us about what you’re particularly interested in?
I’m personally very interested in how people feel when using conversational AI. With various types of conversational AI, including ChatGPT, emerging and continuing to do so, I believe it’s incredibly important to understand how people perceive and experience these systems when they use them.
Through this service, I hope to make conversations with AI more positive and enjoyable for users.
visitors had the opportunity to try out LANGX firsthand.
“The biggest appeal is being able to immediately offer cutting-edge technology to users.”
— What do you think is great about the company?
Since we are a startup originating from a research lab, the ability to engage in both engineering and research is a major strength. One of the most appealing aspects is being able to take cutting-edge technology developed through research and immediately offer it to users.
Being able to work on systems that are 70–80% complete and figuring out how to bring them to 100%, as well as working with the latest technology like real-time conversational AI using CG images, is something I believe you can only experience at this company in Japan.
Additionally, since the organization is still developing in some areas, it’s great that if you have an idea and propose it, you can build a service or workflow from scratch, creating something from zero to one.
“It’s interesting how with each new person who joins, the new culture they bring gradually becomes rooted within the company.”
— What aspects of the company’s culture do you find appealing?
There are system researchers, language researchers, and even people who were English teachers in the past. I find it wonderful that these different cultures come together and are integrated into one system, making it a reality.
Since we have people from diverse backgrounds, it’s interesting to see how the culture evolves with each new person who joins, as they bring their own unique perspectives. It feels like the company’s culture is becoming something positive and dynamic.
However, as a startup originating from a research lab, there’s still room for improvement in the engineering domain. The company is actively recruiting and expanding the team, and the opportunity to help build systems to speed up development processes is incredibly fulfilling.
“What’s important is whether you can enjoy catching up with new technologies.”
— What kind of person is the team currently looking for?
We’re looking for someone who can handle MLOps, someone who is skilled in machine learning and can manage various engineering tasks related to it. We want to work with someone who can leverage new technologies to tackle business challenges together.
On the other hand, conversational AI is still not widely adopted, and many companies, including ours, are experimenting and figuring things out in this field. As such, there are still many aspects of the technology that are not fully established. We try out and implement various techniques during development, so it’s important to have a strong intellectual curiosity, be able to catch up with the latest technologies and information on your own, and enjoy that process.
— Thank you so much!
Equmenopolis is currently looking for new team members to join us!
List of Available Positions
The MLOps/Machine Learning Engineer will collaborate with research scientists to design and implement an MLOps infrastructure that efficiently supports the machine learning model development cycle (e.g., streamlining data collection and annotation, managing training and evaluation, automating deployment and monitoring, etc.), as well as optimizing and speeding up model inference.
We place emphasis on practical experience in designing, building, and implementing machine learning workflows, constructing CI/CD pipelines for stable deployment of machine learning models, and designing and implementing machine learning APIs.
