Falling in Love with “Clean Data”—CAIO Takes on the Challenge at GVA TECH: “Creating New Value in Legal Affairs with AI”
Introduction: From the Recruitment PR Team
At GVA TECH, we explore the possibilities of AI every day to support the future of LegalTech. At the heart of this is the position of “CAIO (Chief AI Officer),” which leads our strategy and development in the AI domain.
In this installment, we interviewed Mr. Iwaki, the CAIO leading GVA TECH’s AI team! From the reality of the still-unfamiliar CAIO role to the behind-the-scenes of AI development and the type of people we want to work with—this interview is packed with hints for anyone looking to get involved in AI x LegalTech.
We hope this gives you a glimpse into the GVA TECH development department.
▶ About the CAIO Role
― Please tell us about your current role and mission.
Yes. My mission is to enhance the convenience of LegalTech services through AI, and I am responsible for technical verification, requirements definition, design, development, implementation, and operational processes for AI-related feature development.
― How do you perceive the position of CAIO (Chief AI Officer)?
When I joined in 2018, very few people in Japan held the title of CAIO, and I felt it had low recognition both inside and outside the company. While the position already existed overseas, it was rare in Japan.
Even now, because the word “AI” is so generic, there is a challenge in conveying what role a CAIO actually plays. Personally, I consider it to be “the person responsible for promoting and overseeing activities that add value to data.” In traditional IT, data usage often ended at input, storage, and search. By using AI, we can derive new insights and discoveries—I truly feel that this is the role required of a CAIO. I want to continue being someone who can share new “insights” both inside and outside the company.
▶ Encounter with GVA TECH
― Could you briefly tell us about your career so far?
From university through graduate school, I conducted research on optical semiconductors. After that, through various twists and turns, I worked as a data scientist on various AI-related projects.
― What sparked your interest in the AI field?
Since I was in university, I was interested in theory and experiments → numerical calculation (simulation) → (???). Twenty years ago, I never imagined that machine learning (AI) would be considered a candidate for (???), but after exploring ways to utilize experimental results (data), I found myself immersed in the AI field before I knew it.
― How did you get involved with GVA TECH?
It started when a former colleague invited me to work at GVA TECH as a side job while I was a data scientist at my previous company. I had previously experienced projects related to natural language processing, but I remember struggling with analysis because most of it was just short “tweets” or “emojis” from social media. On the other hand, when dealing with contracts in LegalTech, I was moved by the fact that contracts are written in “clean and meaningful natural language,” which was the deciding factor in my decision that this was the place to do natural language processing AI.
▶ Current Role and Responsibilities
― Please tell us about your main duties as CAIO.
My main duties involve receiving development requests, conducting technical verification and prototyping, solidifying requirements, and connecting them to actual feature releases. I am responsible for leading the process all the way to the final release.
― Could you tell us about the projects you are focusing on?
All of them (laughs). Although the release dates differ, I am always handling multiple projects, and recently, the focus has been on projects utilizing generative AI. Because the speed of evolution in generative AI is so fast, I am trying and erroring with new technologies and methods almost every day.
― We heard that you use not only generative AI but also traditional machine learning.
Yes, that is exactly right. Generative AI is very powerful, but it is not a panacea for every problem. Depending on the scale and nature of the task, classical machine learning is often more rational and effective. I perform technology selection based on the objective and flexibly choose the optimal method.
▶ Regarding Technology
― What are your thoughts on the direction of GVA TECH’s AI utilization?
In recent AI development and utilization, the high problem-solving capability of generative AI for “forward problems (imitation)” is attracting attention. This trend will likely continue to some extent, and initiatives like using AI to develop AI may accelerate. On the other hand, to avoid the black-box nature of input-output relationships, I believe that responding to “inverse problems”—that is, explaining the rationale of “why it happened”—will be required in the future. No matter how much AI masters the imitation of inputs and outputs as a “forward problem,” it does not necessarily mean it can address the rationale as an “inverse problem.” Therefore, while focusing on development that rapidly converts the progress of AI’s “forward problem” solving capabilities into practical convenience in the short term, I recognize the need to be conscious of efforts toward solving “inverse problems” in the long term.
― What difficulties do you feel in handling AI in the legal field?
When handling highly confidential information like contracts, there are many points to consider compared to publicly available information. The most difficult part is that we cannot visually inspect the specific contents written in our customers’ contracts for product development. In general AI development, we use actual raw data for learning, verification, and test data, and sometimes we strive to improve accuracy while observing the specific data content with our own eyes. However, since we cannot do that, we develop by preparing data that is close to what is expected in actual use and supplementing the missing parts with imagination.
― Are there things you are conscious of in prompt design and LLM utilization?
I am conscious of ensuring reproducibility and suppressing regression (degradation). Basically, later models tend to be superior in output content and cost, but it is not enough to just fix the prompt and adopt a better model. A prompt that was best for a previous generation model might not work well with the next generation model, so when updating models, I strive to ensure output reproducibility or suppress regression through careful prior prompt verification.
▶ About Organization and Culture
― Could you tell us about the composition and atmosphere of the AI team?
The team is composed of an even split between full-time employees and contractors, and it is a full-stack structure that handles not only AI model development but also overall design and implementation, including infrastructure and APIs. We have a diverse group of members with different backgrounds and areas of expertise, and we spend a lot of time on technical discussions. While we sometimes have lively casual conversations, it is a very focused and reliable team.
― What kind of people would you like to work with?
In terms of skills, I want someone who has a strength they can confidently say, 'I am confident in this area of AI-related product development.' In terms of mindset, I would like to work with someone who has a spirit of inquiry and flexibility toward new technologies and who enjoys discussion.
▶ Finally
― As the GVA TECH AI team, what are you aiming for from here on?
We have been, and will continue to be, greedily persistent in our activities to 'create added value in data through AI.' We will continue to take on challenges to create technologies and products that are useful to society.
― Please give a message to your future colleagues!
Let's work hard together and contribute to the development of AI and legal tech. I look forward to meeting those who have curiosity and passion.
― Thank you very much!
At GVA TECH, we are hiring for various positions to realize our purpose of 'eliminating the barriers between law and all activities.' If you are even slightly interested, please let us talk, starting with a casual interview.
↓ Recruitment Site
↓ We started a tech blog in May 2025!
We will continue to increase the number of articles, so please take a look!
↓ Company Introduction Material