Making the “absolutely impossible” possible



Original Source Here

Making the “absolutely impossible” possible, the time of founding AI inside and behind the birth of “DX Suite”

AI inside was founded in August 2015. Since then, the number of employees has increased, and the organization has expanded. We are now promoting our business in a system with more than 30 units. With the mission to bring AI to every being in the world and contribute to a richer future society, we have been working towards realizing a world where AI is used in various things.

In this article, we interviewed our founder and CEO, Toguchi Taku, and Noda (VP of Service Development & Refactoring Unit), who’s been involved in development since founding the company. We asked them questions about the timing of the foundation and behind the birth of our “DX Suite.”

Representative Director, President and CEO, Toguchi Taku (in the image right)

Toguchi began his research and development of artificial intelligence in 2004. He founded the company in 2015 after conducting research and development as well as commercialization and strengthening of financial capabilities. He leads the service development and technical/management strategy of AI inside Inc.

VP of Service Development & Refactoring Unit Noda (in the image left)

Joined AI inside in May 2016 and has been involved in development since then. He is currently the Vice President of the Service Development & Refactoring Unit.

Everyone working on everything with a blind effort, even both development and sales

Comparing 2015 to the time of foundation, what has changed?

Toguchi: Compared to then and now, the organizational system has changed a lot. The current organization structure has the development unit and the sales unit separately, of course. However, when I founded the company, each of us had to develop and sell our product simultaneously. It was usual for us to work in both roles, and we were a true start-up company. Since I’m a programmer myself, I was in development, but at the same time, I was in sales. Noda-san joined as a developer, but he was also in sales in a sense.

Noda: True. Regardless of the position, we were all involved in development and sales, and we’re working really hard.

Toguchi: The entire company was naturally sharing information company-wide on the progress in AI, product development, sales, and issues that our customers faced. Even while being in development, we knew the new users of the “DX Suite.” Everyone had a good understanding of how to use “DX Suite,” so there was a lot of feedback on the product.

Noda: I joined in May 2016, but Toguchi-san was the only developer in the company back then, and I was honestly surprised. I was first involved in development as a freelancer, but after talking with Toguchi-san, I was inspired by the strategy and ideas and decided to join. The day I joined the company was very memorable, I remember being surprised by some of the other employees who were unique.

Toguchi: When I think about that time now, there are quite a few things that make me laugh. There was a time when we wore costumes to the office and had meetings on Halloween.

Noda: Yes, I remember that day, this brings back memories!

What kind of services were you developing at the time?

Noda: At first, we were working on a simple system to input the image, read, and output the data as a CSV file. The service was very simple, like checking whether the characters read on the screen are correct.

Toguchi: Once the PDF is in the folder on the PC, it’s processed in the cloud through FTP*. After that, the CSV is returned to the folder on the PC. Like no-coding software, this was no interface to use.

Noda: This was really groundbreaking at that time!

Toguchi: It was November 2017 when we started providing “DX Suite,” the service currently used by many customers. During the time between foundation until November 2017, we built relationships with more than 500 customers, collected form data, and utilized them for our development.

The current “DX Suite” has continued learning over the years, so the reading accuracy of handwritten words has increased. However, at that time AI-OCR was still something rare, and I think expectations for the accuracy of AI were higher than it is now. Their expectation of AI is that AI can read with 100% accuracy. So if the reading results were even slightly wrong, customers contacted us immediately.

Noda: At that time, there was an expectation that “AI is perfect and will not make any mistakes,” so we needed to educate our customers on what AI-OCR is. Another word of saying, we were not just selling them, but we were also spreading the understanding of AI.

Toguchi: I agree. I think we’ve been working on making society recognize what AI-OCR can do.

*FTP: abbreviation of File Transfer Protocol. Communication rules for sending and receiving files between a server and a client.

Create a prototype in 1 week and propose to the customer

How did you come up with the current DX Suite?

Toguchi: While we provided our service to our customers, we received many requests from inside and outside the company. To find a service that will fulfill customers’ needs, we worked on a design sprint on a whiteboard. We prepared samples of the forms and tested ourselves to see where users were taking too much time.

Noda: This is when we made improvements on the identified issues, and this one was really close to the current “DX Suite.”

Toguchi: Yes, but just a part of it. Up until this point, it took about one week. We used the first day for the design sprint, the second day to finalize the requirements with design drafting, and the third day to create it. On the fourth day, we tested and fixed the issues on the fifth day.

Noda: Back then, we were all collecting data for validation, training, and working on annotation* at the same time. That was a lot of work.

Toguchi: It really was. We all just worked in silence and created data. I didn’t realize this back then, but looking back on it later, I think that data was valuable as a treasure.

Noda: If I recall correctly, our first order was for a case Toguchi-san was in charge of.

Toguchi: The first order we had was in August 2017. Until this, we didn’t exactly have a service name for it, but this is around the time we named this service “DX Suite.”

A future would certainly come where DX will progress in the world, so we named it with the idea of creating a service cluster to support DX. The word “Suite” means to layer, and the intention we had was to layer various services within a single software package.

This service started with the concept of AI-OCR “Intelligent OCR” as a function on “DX Suite,” not just a single service to use AI-OCR. In the beginning, we only had “Intelligent OCR,” so it was not actually a “Suite.’’ Now, we have other optional features aside from “Intelligent OCR’’ such as the automatic sorting function “Elastic sorter”, and users can select optional functions to suit their business needs. In addition, we already had an intention to take in AI that is created with partners and other companies on our service. This idea links with the platform strategy or the marketplace that we are currently aiming to achieve.

Noda: After the service release, we found that many areas were not taken into account, and we have continuously fixed and made improvements. Everyone in the company had an account for the “DX Suite,” and everyone tested their service.

Toguchi: If we didn’t use our service ourselves, we wouldn’t be able to propose them to our customers. What we were doing was completely dogfooding*.

Annotation: To create training data (ground truth data, labels) to train the machine learning model.

Dogfooding: To test the company’s products and services by employees in the IT industry, etc., while using them in actual ways on a daily basis.

A breakthrough that made the “absolutely impossible,” possible

How was the year after releasing “DX Suite” like the year 2018–2019?

Toguchi: It was a fruitful year for us as we expanded our services, worked on projects with other companies, and increased our alliances. This year was also around the time the accuracy of the reading improved.

Noda: The accuracy rapidly improved in May 2018. One of our engineers came up with this idea while he was on a workcation, and it led to a breakthrough in improving the accuracy of AI-OCR.

Toguchi: I think it was around August of that year when the number of contracts acquired per week started to increase. This was when we set up the customer success department, and it was during this period when we began to improve the quality of our services and create a follow-up system for our users. Ever since then, we have been enhancing our system.

Noda: We also had a breakthrough related to the “Elastic Sorter.” In those days, we were trying to implement OCR for sorting forms, and we were wondering what kind of UI/UX or functions should be developed to handle the complex settings and operations. That’s when I discussed with Toguchi-san, and we were able to realize this by “sorting with feature*.”

Toguchi: It’s very “ordinary” to consider sorting by OCR. If you fail, you will know why you failed, and the user can control how they want to process. However, in my opinion, no user wants to take that process. I thought that users are seeking improvements in operations beyond sorting. This is why I proposed an idea that doesn’t require any setup for sorting at all.

I remember the people around me saying that the sorting feature was “absolutely impossible” and “no company does that.” If that’s the case, it just gives meaning to do it, and it was a good chance and opportunity.

Noda: Well, we were able to realize this after all. Now, just by uploading the form data to “Elastic Sorter”, you can identify hundreds of types of forms and sort them instantly. This is a difficult task for people to do, so users are very happy with this function. Also, by applying the technology of image recognition by features, we are now able to automatically correct angles of the tilted documents, and read texts on forms with even more patterns.

*Feature (value): Indicators that become patterns for recognizing images.

Shared and innovated by all

Lastly, what kind of organization do you want to aim for in the future?

Noda: As our company grows, new talented members are constantly joining. Still, as a startup company, I’d like to keep the mindset of taking on challenges ourselves.

Even when faced with difficulties, we have the persistence to gather everyone’s wisdom and make full use of the options. I believe this is the reason we’ve been able to make innovations to this day. To achieve this, we need to challenge efficiently in a blind effort.

As we aim to become the number one company in the world, I’d like to continue cherishing the stance of moving forward while sharing every moment. I’d also like to enjoy the chaos even more too.

Toguchi: One of the reasons why AI inside has continued to grow is that we’ve been innovating in things people said we could never achieve. I’d like to continue to emphasize the importance of our corporate culture in which we all work together to steadily address issues, make improvements, and move forward.

For the “Rules and Innovation” which is a mindset that we value, please see here.

AI/ML

Trending AI/ML Article Identified & Digested via Granola by Ramsey Elbasheer; a Machine-Driven RSS Bot

%d bloggers like this: