Content

Introduction

Foundations

System Architecture

Artificial intelligence is rapidly becoming part of children's everyday lives. Whether they are asking questions, learning new skills, exploring their interests, or seeking emotional support, children are increasingly turning to AI as a trusted source of information and conversation.

Building AI for children, however, is fundamentally different from building AI for adults.

Children are still developing emotionally, socially, cognitively, and morally. They communicate differently, understand context differently, and often place a greater level of trust in technology. These differences require more than simply filtering inappropriate content or simplifying language. They require AI systems that recognize the unique responsibilities involved in interacting with a child.

Sisbot was created to explore what a truly child-aware AI should look like.

Rather than treating child safety as a collection of isolated moderation rules, Sisbot approaches safety as a system-wide design principle. Every component—from personality design and conversation flow to memory, reasoning, tool usage, and model orchestration—is intentionally designed to support children's wellbeing while respecting their curiosity, autonomy, and capacity to learn.

This documentation is a comprehensive record of that work.

It explains not only what we built, but why we built it, the principles that guided each decision, and the engineering choices that transformed those principles into a working system.

Throughout these pages, you will find:

• The philosophy behind child-aware AI and the challenges it seeks to address.
• The research and design principles that shaped Sisbot's behavior.
• The adaptive guardrail framework that enables context-aware safety instead of rigid rule enforcement.
• The personality model that balances warmth, trust, and healthy boundaries.
• The system architecture, including prompt design, memory, model routing, and tool integration.
• The evaluation methods and continuous improvements used to assess safety, reliability, and user experience.

This documentation is intended for:

• AI engineers,
• Product designers
• Researchers,
• Educators,
• Child safety specialists,
• Non-governmental organizations,

And anyone interested in building AI systems that place children's wellbeing at the center of their design.

<aside> 💡

Sisbot is not presented as the definitive solution to child safety in AI. Instead, it is an evolving framework—one that combines research, engineering, and practical experience to explore how AI can become not only more capable, but also more responsible.

</aside>

As artificial intelligence becomes increasingly integrated into children's lives, the question is no longer whether children will interact with AI. The question is how responsibly we choose to build it.

We hope this documentation contributes to that conversation and encourages others to design AI systems that are safe, trustworthy, transparent, and genuinely supportive of children's growth.

Welcome to Sisbot.