I Built an AI Stretching Coach for a Kids' Speaker (Yoto)
Last month Yoto announced a developer challenge for their screen-free audio player, and I thought "why not?"
So I built Daily Stretch - an app that generates a fresh stretching routine every single day. No screens, just audio guiding you through 5-10 minutes of movement.
The Setup
The tech stack is pretty straightforward:
- Claude writes the daily stretching scripts
- Each script is for a specific day, using a monthly theme (for example, October is fall + Halloween + fun spooky)
- Gemini Text-to-Speech turns them into audio
- API endpoint just points to the right file (deployed to Google Cloud Run)
The Tricky Part
Getting the AI to generate good stretching instructions was harder than I expected. The first few attempts were... not great. Claude would rapid-fire instructions without giving you time to actually DO the stretch (and it still does this on occasion).
It took a few iterations to get the prompt right. I had to be really specific about:
- Building in pauses between instructions (you need time to get into position!)
- Timing each hold properly
- Smooth transitions between stretches
- Keeping an encouraging, calm tone throughout
Turns out there's a big difference between "generating a list of stretches" and "creating an audio experience that feels natural."
What's Next
Daily Stretch is live! I'm curious whether people will actually use it daily or if the novelty wears off after a week or two.
If you have a Yoto player and want to try it, I'd love your feedback - especially if you use it with kids. Does it hold their attention? Is the pacing right?
I'm also thinking about this pattern more broadly - AI-generated daily content that feels natural and human. Could work for other types of audio experiences too.