
Personal Project
An LLM designed specifically for children. Friendly characters, age-appropriate guardrails, full parental controls. I designed it properly — character-based chat interface, multiple AI personalities to interact with. Then simplified it for real testing. Built a working version with push-to-talk and text input. Then I tested it with my own kids.

Design 1
Lost the developer halfway through. Rather than abandon it, I decided to vibe code it myself. Stripped everything back to basics. No character system. No fancy animations. Just a text box, a microphone button, and an AI that wouldn't say anything inappropriate to an 8-year-old. Two months of evenings and weekends. Mostly worked.



Tested it with my own kids. They were excited for about five minutes, then went back to asking me instead. That was the real insight. Kids are endlessly curious — my son asks me things I have to look up constantly. But they don't want answers from a screen. They want answers from someone they trust, someone with experience, someone who can read the room when the answer needs simplifying or expanding. An AI can answer the question, but it can't be Dad. The other mistakes were more straightforward: (1) web-based was dead on arrival — they live on iPads, not browsers, and (2) parents aren't actively looking for 'safe AI for kids' yet — they just limit screen time. The concept was technically sound. But I was solving a supply problem (safe answers) when the real dynamic was about trust and relationship. That's the kind of thing you only learn by testing with real people — never from a design tool.