Leon Xu joined JAPAN AI in October as the Head of Product Engineering, bringing with him a career spanning backend development, infrastructure, DevOps, and engineering leadership across industries such as recruiting, news, and gaming. Over the years, he has built and scaled engineering organizations from the ground up, worked side-by-side with ML teams, and led multiple product teams through periods of high uncertainty and rapid growth. He has also explored AI on a personal level, prototyping his own RAG-based chat systems and experimenting with how large language models can be grounded in real-world data. Today, he leads JAPAN AI’s product engineering organization—an essential group responsible for building a fast, scalable, and high-quality foundation for all JAPAN AI products. We spoke with Leon about why he joined JAPAN AI, the challenge of balancing speed and quality, and what it means to build an AI platform designed for the future rather than just the present.
Leon Xu
Product Dev Group
General manager
I’m Leon, and I joined JAPAN AI in October as the Head of Product Engineering. Throughout my career, I’ve mainly worked as a backend, infrastructure, and DevOps engineer across domains such as recruiting, news, and gaming. That variety of industries exposed me to very different types of systems—from latency-sensitive services to data-heavy backends—and taught me how to design for both reliability and change.
In the past few years, I’ve taken on roles that required me not only to ship features, but also to design and grow the engineering organization itself: hiring engineers, defining development processes, and aligning technical decisions with product strategy. In that sense, what I do at JAPAN AI now is a natural extension of what I’ve been doing—just in a much more dynamic AI-driven environment.
My involvement in AI began when I collaborated closely with ML teams on news recommendation systems. That experience showed me how much value can be unlocked when traditional backend systems and ML models are properly integrated, instead of sitting in separate silos. Later, I led a team focused on content generation and content understanding using LLMs, which gave me hands-on experience with prompt design, evaluation, and the operational challenges of running LLM-powered features in production.
On the personal side, when GPT-4 came out, I experimented with building a RAG-based AI chat prototype on my own. I was curious about how far you could go with retrieval-augmented generation without building a massive infrastructure around it. Those experiments convinced me that AI would fundamentally change how we design products and systems.
AI has always been a natural extension of my interests and experience, so joining JAPAN AI felt like the perfect next step.
The AI field is evolving at an unprecedented pace. Startups everywhere are attempting to transform business models and industries through AI, and almost every company is trying to label itself as “AI-driven” in some way. In that kind of environment, it’s easy to get lost in hype.
What stood out to me about JAPAN AI is that the company isn’t just building another vertical solution on top of AI—it’s building a horizontal AI platform. Instead of focusing on a single industry, JAPAN AI aims to provide a foundation where AI agents can power many different workflows and use cases across multiple sectors.
We don’t develop foundational models like OpenAI, Google, or Anthropic do. Instead, we focus on what they can’t easily do: deep system integration with enterprise environments, alignment with the realities of Japanese businesses, and building products that remain valuable even as the underlying models change. That combination of platform thinking + local insight is rare.
For me, that’s not just an interesting business angle—it’s a survival strategy in a fast-moving field. In a world changing this fast, building a horizontal platform isn’t a “nice to have”; it’s essential if you want to create solutions that stay relevant over time. That long-term, strategic vision is what made me want to join JAPAN AI.
The most memorable experience so far has been the release of JAPAN AI STUDIO.
From the moment I heard about JAPAN AI STUDIO’s concept, I was excited. It’s a product that doesn’t just expose models or APIs—it provides an environment where agents, tools, and workflows can be composed in a way that feels natural to both developers and non-developers. In that sense, it is genuinely ahead of some major players.
But turning this concept into a real, usable product required solving a classic but very hard problem: how do you move fast without breaking everything? How do you ship something ambitious at high speed while keeping the quality bar high enough that customers can rely on it?
When I joined, JAPAN AI STUDIO was entering a critical phase. There was a lot of great work already done, but also several gaps: