The core insight: filtered adult AI isn't the same as AI built for children. A content filter catches bad words but doesn't adjust vocabulary for a 5-year-old vs. a 12-year-old, or know when to redirect a conversation.
Askie is voice-first (kids talk, not type), COPPA compliant, and gives parents full conversation transcripts. Image generation is built on Google Gemini with child-appropriate outputs.
Tech stack: Google Gemini for voice/conversations, Google Gemini for image generation, Railway for hosting, MongoDB, RevenueCat for subscriptions.
Currently serving 2,500+ families. Would love feedback from the HN community on the approach to child safety in AI.