At the AI Expo held in Korea, I saw a platform called https://www.soak.so/en/main (I have no conflicts of interest) which seemed like such a safe and beneficial application that I thought if I had a child, I would want to educate them with it. I strongly recommend it.

I believe they mentioned they were developing an LLM-based question-answering feature, but it doesn't seem to have been implemented yet.

You can prob build something pretty simply using this (it sits between user and the llm and intercepts unwanted content):

https://platform.lakera.ai/

This can be achieved with a thorough system prompt. Would take no time to make if you are willing to pay a freelancer.

In my limited experience you can ask some (I tried with chatgtp and gemini) llms to prompt themselves. Both came up with good prompts that achieved the goals we had in mind.

How wiki could be the threat here?

There are topics, pictures and instructions that are not suitable for 11 year olds but are for adults. Consider the wiki entries on certain matters of extreme violence, ideology or adult culture.