Conversation between Machine Learning Engineers. Meet @ lain21 and @ chica #WeMakeMercari
Disclaimer: This interview is a translation of an interview conducted in Japanese, which was conducted before working from home started. All Mercari employees are currently working from home as stated in our CEO statement in regards to the novel coronavirus.
In the #WeMakeMercari series, we would like to introduce you to some of the remarkable members of the company. In this episode we have Lain Matsuoka (@lain21), a Machine Learning Engineer from the US@Tokyo ML/Data Team, and Chika Matsueda (@chika), also a Machine Learning Engineer from the Mercari Edge AI Team. In this article, we will discover more about their professional background and learn what’s special about AI teams in Mercari.
Nice to see you again!
@chica：Hi Lain, long time no see!
@lain21：Hi Chica, nice to see you again! I moved to the Mercari US Team in Tokyo (US@Tokyo), so I’ve been losing touch with you recently. How have you been?
@chica：In my free time I started a new club channel on Slack called #z-beauty. We actually talk a lot about fitness! I started going to personal training sessions, and the first week I tried a low-sugar diet, but it didn’t last so long. But then I changed to a low-fat diet and I started getting results.
@lain21：One day I would like to participate in the Médoc Marathon in France, where you drink good wine and run at the same time. I will have to prepare my body but especially my liver for it! (laughs)
From left: @lain21 (ML Engineer, US@Tokyo ML/Data Team) and @chica (ML Engineer, Mercari Edge AI Team)
@chica：Seriously? A marathon and drinking wine event? I didn’t know there was such a thing. By the way, I know you graduated from the pharmacy department at your university. Please tell me more about your background, I’m curious.
From studying Pharmacy to Space Technology: The path before Mercari
@lain21：Okay, so where do we start? After graduating from the pharmacy department at my university, I went on to graduate school to continue my studies in the same area, but I ended up dropping out. Then I became interested in space technology after reading the comic book Space Brothers… Dreaming about working at NASA, I did a one-year internship at JAXA (Japan Aerospace Exploration Agency). After that, I went to the US to study aerospace engineering in graduate school. Completing graduation, I joined the Massachusetts Institute of Technology (MIT) Aerospace Engineering Doctoral Program, researching applications of machine learning in AI robots, but I dropped out after a year.
@chica：You have dropped out twice, so maybe you have a tendency to get bored after a year (laughs).
@lain21：Maybe you are right (laughs). After returning to Japan, I participated in various internships at Mercari and other Japanese companies as a machine learning engineer. Then, I formally joined Mercari JP as a machine learning engineer in January 2019.
@chica：I remember the AI Team had about 12-13 members at the time. As an ML engineer myself, I remember being very happy to know a female member had joined the team.
@lain21：When I was in Mercari JP I created the price estimation model, but after about two months I moved to the Merpay Machine Learning Team. Then, I transferred to the US@Tokyo Team in January 2020.
@chica：Wow, so you experienced working in three teams in one year! Would you like to work in the US eventually?
@lain21：Yes, I hope so.
I decided to become a game programmer when I was an elementary school student
@lain21：How about you, Chica? How did you become an engineer?
@chica：Actually I have liked video games since I was in elementary school. So, I decided to work in the video game field when I grew up. There are many areas in the video game field like a game programmer and game designer. I tried all of them when I was in elementary and junior high school, and decided to become a programmer as it was the most fun among them.
@lain21：Wow, you had a clear vision of what you wanted to do at a very early age!
@chica：Then I joined the Faculty of Information Science at a women’s university. There I met girls who liked numbers, but I didn’t meet so many girls who liked programming. I considered myself to be someone who knew about programming, but when I was a 4th-year student I participated in the Google Women’s Internship BOLD and for the first time I met women who were really good at programming. I felt like a small fry put among big fish (laughs).
@lain21：You are not a small fry! (laughs)
@chica：From that experience I felt really motivated and studied very hard. After that, with tenacity I was able to make my childhood dream of joining a video game company come true. But then I realized something. I like playing video games, but I don’t like making them that much…
@lain21：What?! Why did it take you so long to realize that? (laughs)
@chica：The serious part of video game programming was interesting, but I ended up quitting in two years. Then, I joined a friend’s startup company which specialized in data analytics. I wanted to build up experience in data analytics, but the company was sold to another company and the atmosphere changed completely.
@lain21：Did you join Mercari after that?
@chica：At the time, Mercari’s BI Team was doing a 1-Day Candidate Selection for data analysts, which was a set of various interviews and the whole selection process at once. It looked interesting so I applied, but sadly I didn’t pass. However, the interview was pretty interesting so I started to learn more about Mercari. Then, I found an open position for an engineer in the AI Team, I decided to apply, and this is how I am here now!
A Python lover going to an unexplored area
@lain21：Chica, I know you are a Python lover (an essential for ML).
@chica：Yes, true. I became a Python lover when I was in my last year of college. At that time, I was doing an internship and my manager suggested I lacked basic knowledge so I should make myself a master in something. Then, I decided to become a Python master. I thought that becoming a Python master I would be able to do other things, so I read all the documentation related to it. And then, I became able to estimate how much I would be able to develop in Python.
@lain21：As a programming language, Python is just the right scale. It has many useful features and you can do many things with it.
@chica：Also Python is great for someone starting to study programming. Before it was not the main language used for development, but still it was a very useful tool. It’s like talking about Go now. For me at the time, using Python was similar to supporting an underground artist (laughs).
@lain21： I haven’t used Python to write code recently. I only use PHP and Go. But tell me, what were you doing after joining Mercari?
@chica：I joined the AI Team at the time, getting basic knowledge while working on Sys ML, Kubernetes and refactoring and optimizing existing violation detection systems. Then from October 2019, I transferred to the Edge AI* Team working on developing an Edge AI technology base. The team’s mission is to implement an existing model on smartphones without creating a new machine learning model. But I cannot find my way out…
* Edge AI refers to the technology to run AI algorithms locally on a hardware device using edge computers.
@chica：TensorFlow Lite* and MediaPipe* are still in the development process so even if we use it as it is, it doesn’t work most of the time. We need to fix the existing code in order to make them work. However, I’m not a complete expert in ML and deep learning, so I haven’t been able to find an answer yet, and I cannot find my way out.
*TensorFlow Lite is a set of tools to help developers run TensorFlow models on mobile, embedded, and IoT devices. It enables on-device machine learning inference with low latency and small binary size.
*MediaPipe is a graph-based framework for building multimodal (video, audio, and sensor) applied machine learning pipelines.
@lain21： I totally understand how you feel. Even if I send a pull request it won’t be merged.
@chica：Exactly! However, I feel happy that for the first time in my life I’m trying something that no one has done before.
@lain21：That’s amazing! I didn’t choose my career path with such motivation. I just entered MIT because there was someone who I fell in love with at first sight (laughs).
@chica：That’s also a good story!
@lain21： By the time I joined MIT that person had a partner already (laughs).
Smart Pricing: a characteristic feature of Mercari US
@chica：Recently, Mercari JP adopted a camp system, which basically connects teams by function. In the Edge AI team, we are in charge of the platform, but actually there is no platform for Edge yet. We don’t plan to make a platform all at once, but we are trying to collect and use study cases and knowledge. Currently, I’m working on real-time price estimation, but it’s interesting to be able to develop new features from the bottom up in the camp system.
@lain21：I think the Edge AI Team is close to the work of Research and Development (R&D).
@chica：Indeed! How about you, Lain? What are you doing now? Mercari US has a totally different UI from Mercari JP, and I guess it has some item particularities, right?
@lain21：In Mercari US we have a feature for automatic price discounting called Smart Pricing. If the user activates this feature, the system will automatically lower the price by 5% each time until it reaches the lowest price set in advance. Also, it brings the listing to the top of search results every time the discount is activated. The Smart Pricing feature obtained quite good results when it was released, so now we are working on improving its algorithms.
@chica：Wow! We should have that function for Mercari JP as well!!
@lain21：I’m also in charge of implementing the ML model developed by Mercari US in the production environment with microservices. I’m really doing a lot of things such as data analysis, thinking of new ideas and even if it’s just simple logic I implement them on the backend, etc. We work closely with our colleagues in the US, so it’s pretty common to have meetings from 8 am Japan time, and that is a bit tough sometimes (laughs).
The characteristics of the AI teams in Mercari JP, Merpay and US@Tokyo
@chica：So you have experienced working with all the AI teams in Mercari JP, Merpay, and now Mercari US. Could you tell me about the differences between them?
@lain21： In the Mercari JP team I was able to take my time to work on tasks. I was able to try things without worrying about making mistakes, and I was able to do research and take time for studying too. I had to because I was not able to put the machine learning measures into production without knowing about the backend and infrastructure.
@chica：I see! How about Merpay and US@Tokyo?
@lain21：In Merpay I had much more pressure and a sense of urgency than in Mercari JP. Naturally, Merpay is a financial service and we had to create a good system that supported everything related to payments. I was also involved in the development of the deferred payment system, and we spent a lot of time talking about credit. On the other hand, in Mercari US there are many things we still need to work on, so you can still feel the startup company environment. It feels like I’m running and stepping on mines while fixing things. Now I’m fully immersed in the product and it’s fun to implement the features I created.
@chica：I’ve always been in Mercari JP, but when I joined the AI Team had about 12–13 members. Later, we had a group of new grads joining from India and elsewhere, so now we have about 50 people, from both Japan and overseas. I realized how my role has been changing in the process. There are many outstanding members and I know I cannot beat their level of expertise so instead, I focus on proper development. I want to support the new grads and help them realize what they want to create. For this purpose, recently I took the Scrum Master Certification. Now I can feel the project speed is improving in each Scrum meeting.
@chica：I feel like I’m not essential anymore (laughs). But I’m glad I was able to create the current environment. How about you, Lain? What is your motivation to work?
@lain21：Well, I think at first there is only personal motivation, so for example, feeling happy with the people I’m working with. US@Tokyo has a great working environment. It’s easy to implement ideas like in a startup company environment. This also means the adrenaline of having to succeed!
@chica：That really sounds like a startup environment.
@lain21：I feel like I’m working the most now since joining the company (laughs).
@chica：I also want to work peacefully in an environment that doesn’t lose the startup spirit! (laughs).