Endeavor Navi

Inspiration

Most AI services require users to be 13 or older, with parental consent needed for teens up to 18. However, many elementary school children are already using them, causing significant anxiety for parents concerned about exposure to inappropriate content. As fathers of young children, my teammates Yasu and Hayato believe in AI's positive potential but recognize the dilemma it presents for kids. We are inspired to solve this by creating a safe, controlled AI learning environment. Our vision is to empower children to explore their curiosity and develop essential AI literacy, while a robust monitoring system gives parents peace of mind. Our goal is to prepare the next generation for an era of human-AI collaboration.

What it does

Endeavor Navi is a "Parental Control AI" application designed for elementary school students. It combines an AI-powered learning assistant with a comprehensive parental control system. The core features include: Safe Dialogue Navigation: The AI filters out inappropriate content, using age-appropriate language to guide conversations. "Insight" Notifications for Parents: When a child explores a topic that may be too complex or sensitive for their age, the system instantly notifies the parent's smartphone, creating an opportunity for a family conversation. Parent-Led Learning Controls: From the notification, a parent can permit or deny the line of questioning with a single tap. Denying a query temporarily suspends the device access, ensuring the child stays on a safe learning path directed by the parent.

How we built it

Using W&B Weive, Google ADK, Gemini as LLM, MCP Native Application Features: Keylogger function Screen lock feature Control by an ambient agent Role of the Monitoring Agent: Monitoring of contextual information Detection of meaningful sentences and words Policy-based determination This project was driven by one engineer and one business-side member. The engineer built the backend, the AI agent, and the demo environment. As the business-side member was responsible for creating the PRFAQ (Press Release/Frequently Asked Questions), the SOW (Statement of Work) for the product development policy, and the prototype for the front-end design.

Challenges we ran into

Integration of the agent and native environment Implementation in physical/local environments Implementation of contextual understanding: Determination based on contextual understanding, not just keywords Context analysis utilizing LLMs Achieving fine-grained control through agent role-based division

Accomplishments that we're proud of

Integration of the agent and native environment as Ambient Agent that is able to understand context Given our direct experience as parents of the target child users, we committed to a user-focused development methodology, ensuring the user journey remained central to our efforts.

What we learned

Gaining practical insights for AI agent implementation in real environments The importance of implementing AI agents closer to human environments The significance of initial planning in AI projects, the necessity of requirements definition and alignment across teams

What's next for Endeavor Navi

Our vision is to evolve Endeavor Navi into a comprehensive educational platform that teaches children to collaborate with AI effectively and ethically. We will achieve this by creating structured learning modules, enhancing the AI's teaching capabilities, and offering parents detailed insights into their child's learning patterns and interests. Endeavor Navi is currently a parental control tool designed for parents and children to support them when using AI services. However, by leveraging its keylogger and approval system features, it becomes possible to implement a system for corporate and organizational compliance, as well as to assist and support team members who may have an incomplete understanding of certain tasks. Furthermore, since it can be deployed in a local environment, it is also applicable for work with strict restrictions and where privacy is a key consideration. We are planning to offer this functionality as a feature for agents through an MCP server.