Delait // AI magic for children (and parents)

Inspiration

Kids need activities that leave the screen, but still benefit from AI guidance. We wanted one place where voice, text, camera, images, and music help create real-world play—under strict parent control.

What it does

Delait runs offline-first games: the phone gives short prompts and check-ins, while the main play happens away from the device (drawing, word games, rhythm, movement, quests). Kids talk or type; Gemini analyzes input and picks a safe activity mode, with optional image/music generation and full parent visibility of everything created.

How we built it

A web app built with PHP/Symfony + vanilla JS, mostly vibe-coded for speed. We designed modular “game modes” (chat, drawing, rhythm, screensaver/landscapes) and a parent dashboard with approval, time windows, and custom safeguards.

Challenges we ran into

The only real challenge was time and money. I built the project after hours using my own funds, so I had to keep the scope under control. An app this large and complex would need much more infrastructure and technical capacity—I have a clear plan for how to scale it. Once expanded, it could be genuinely impressive: dozens of AI components working together as one coherent system. But by my estimates, it would take well over 1,000 person-hours (and I’m just one person army).

Accomplishments that we're proud of

A single app that combines voice/text, camera-assisted drawing feedback, and image + music generation while still pushing kids into real-world play. A challenge is describing everything I’ve already built, while also conveying what’s still in my head—and what I simply didn’t have time to implement yet. And we definitely didn’t have enough time for proper testing.

What we learned

The biggest takeaway is that the AI world and the real world can blend together in a way that feels almost invisible—and that multimodality is the most important capability of AI models. I couldn’t use everything in such a short time (the app was built after hours), but Gemini’s possibilities are genuinely impressive: analyzing audio, images, and text at the same time. I didn’t even try video analysis (yet), mainly due to technical constraints.

What's next for Delait // AI magic for children (and parents)

We need real-world testing with children and time to debug practical issues. Using agents helped us invent games that seem interesting and engaging, but this has to be validated through real trials. We also need collaboration with child psychologists to make the safeguards even stronger. Parent reporting should be improved so kids benefit more from it. Maybe we should add activities that involve parents too. The idea is too big to fully deliver in such a short time.

Built With

• flux
• gemini
• javascript
• minimax
• php
• stimulus
• symfony