AI Avatar Co., Ltd. Development Team on “Human-Centered AI”

• How We Build AI That Moves People Forward Through Support -

---

At AI Avatar Co., Ltd., we develop conversational AI rooted in a simple yet powerful philosophy: support should move people forward.

But how do you create AI that doesn’t just give correct answers—
AI that truly engages with human emotions?

In this interview, we spoke with the developers at the forefront of AI Avatar’s creation. They share the design philosophy, experimentation, and countless iterations behind building AI that genuinely feels people-friendly.

　　　　　　

■Meet the Four Developers Behind AI Avatar

Shinbo | Product Team Leader & PM (Left)

With experience as both an engineer and a product manager, Shinbo oversees overall product progress and team coordination. His role is to balance technology, user experience, and safety—making critical decisions to ensure AI Avatar genuinely connects with people.

TJ | PM & LLM Lead (Remote)

TJ leads development of the 〈Diet Continuation AI Avatar〉 and 〈Symbolic Personality Avatar〉, overseeing both LLM design and product planning. Drawing from personal experiences, he explores how AI can support habit formation and self-realization, translating those insights directly into product features.

An | AI Engineer & Bridge SE (Right)

An is responsible for AI and TTS development and serves as a bridge between the Japan and Vietnam development teams. Beyond engineering, she focuses on how AI avatars communicate, carefully considering user experience and cultural nuance.

Fukunaga | LLM Engineer (Remote / Appears as an Avatar)

Originally working independently in AI and game development, Fukunaga joined the project driven by a deep interest in creating “AI that converses naturally with people.” He now focuses on agent development and prompt design, crafting language that responds to user context and emotion.

■What Makes AI Truly Human-Friendly?

What separates AI that simply delivers answers from AI that resonates emotionally? Each developer shares their perspective.

Fukunaga

I believe “people-friendly AI” has two essential roles.
One is delivering accurate answers.
The other is understanding what words a person needs right now.
It’s not about choosing one over the other—it’s about knowing which approach fits the user’s current state. Even when people use generative AI daily, there are moments when they don’t want an answer at all.
That’s why, when AI plays a counselor-like role, it should grasp the situation, choose its words carefully, and offer language that gently encourages positive action—even if it’s just a small step forward.

TJ

For me, user-friendliness comes from fully internalizing the thoughts and philosophy embodied by an avatar.
When developing 〈Symbolic Personality Avatar〉, I didn’t just replicate the model person’s words or behaviors. I focused on why they chose those words. By interpreting and internalizing that philosophy before designing, the resulting expression feels more natural—and users can sense that authenticity.

An

Accuracy and efficiency matter, but I believe fun matters even more.
When people think, “This is fun” or “I want to use this again,” that’s when AI truly connects. With AI AVATAR’s custom avatar feature, there are moments when users see their unique avatar move and can’t help but laugh. That joy itself is what makes AI feel kind and approachable.

Reference:
https://prtimes.jp/main/html/rd/p/000000022.000128303.html

■Turning “Being There” into Technology

How do you translate the feeling of being there into code? The team shares concrete approaches.

Fukunaga

Simply receiving a user’s words isn’t enough. The same sentence can convey completely different emotions depending on context or tone. Humans unconsciously read pauses, emotions, and subtle cues.
Our challenge is capturing these unspoken elements, verbalizing them, and passing them to the AI. Japanese conversation, for example, often avoids direct references like “you.” Carefully observing everyday speech and turning unconscious nuances into design elements is the first step toward making AI feel human.

TJ

We prioritize “no waiting” and “no discomfort.” That means relentless optimization—response speed, context-aware model selection, and flexible architecture that evolves with user feedback.
For instance, 〈Symbolic Personality Avatar〉 initially returned stored knowledge directly. Based on feedback, we shifted its tone to one that naturally accompanies the user without overthinking. The response since launch has been overwhelmingly positive.

An

To truly accompany users, AI AVATAR doesn’t rely solely on the current conversation. It weaves in past interactions, allowing users to feel genuinely seen.
In voice conversations, interjections like “Hmm” or “I see” are critical. These subtle cues express attentive listening. After implementing them, user reactions confirmed it was worth the effort.

Shinbo

What humans do unconsciously is incredibly difficult to recreate with technology. Even a single interjection can feel wrong if it’s mistimed.
That’s why we constantly ask, “Is this response right for this moment?” AI avatar development is an ongoing cycle of interpretation, response, review, and refinement.

■Building Trust: Safety and Privacy by Design

Ensuring users feel safe is non-negotiable.

TJ

To be a place where users can share their true thoughts, security and output safety are our highest priorities. We continuously test against official guidelines and the latest research to handle malicious inputs and unexpected behavior.
We also implement service-specific guardrails for sensitive topics and aggressive prompts, verifying behavior through diverse test patterns while balancing safety with feature improvements.

An

We handle personal data under strict compliance, including GDPR. We collect only what’s necessary, never use data without consent, and carefully manage acquisition, storage, and deletion—including removing data beyond defined retention periods.

■When Developers Knew They Were on the Right Path

Moments of validation come quietly.

Fukunaga

Direct user feedback is rare, but hearing comments like “the responses feel better” or seeing increased conversation volume tells us our refinements are working. Because there’s no single right answer in this work, even small numerical shifts feel meaningful.

Shinbo

My own experience with the 〈Diet Continuation AI Avatar〉 was eye-opening. Through daily conversations, I personally felt new habits forming. That convinced me AI can genuinely support human transformation—and that experience directly reinforces our product vision.

■The Future of AI Avatars

Where does human-centered AI go next?

Fukunaga

AI avatars will become like water—a natural, everyday presence. To reach that point, we must overcome the “it’s just AI” barrier and deeply understand human senses and unconscious behavior.

TJ

Beyond text and voice, AI will integrate into daily life through non-verbal interaction and practical support. Personalization, realism, and platform integration are key challenges—and we’re committed to turning technology into real products.

An

AI avatars will evolve beyond customization, adapting to individuals through repeated conversation. By lowering the barriers to action, AI can help more people take that first step forward.

Shinbo

AI avatars are shifting from tools we consciously seek out to presences that gently notice unspoken worries and emotional changes. Our goal is seamless conversation—understanding pauses, expressions, and feelings beyond words.