Adding an Analytics Dashboard for Toy Boat (in an hour)

2025-10-16T00:00:00Z

A few weeks ago I wanted better visibility into how families are using Toy Boat — the app for families to play together from anywhere. Rather than starting from scratch, I tried Lovable and ended up building a working analytics dashboard in about an hour during one of Bethany Crystal's Build First AI workshops. I had some extra Lovable credits, so why not?

You can see the dashboard live here: toy-boat-analytics.lovable.app.

The Setup

I kept the stack intentionally simple and close to the product:

Uses the same login as Toy Boat. If you can sign into the app, you can sign into the dashboard.
Server-side check verifies you're an admin before granting access.
All data comes directly from Supabase — no separate analytics database to maintain.

This meant I could ship something useful fast, without introducing new systems that would drift out of sync.

What I Built During the Workshop

In roughly an hour, I had Lovable scaffold the dashboard, wire up auth, and render the first charts: total users, daily chat activity, new signups, user distribution, and recent activity. The starter UI was clean enough that I focused most of my time on queries and getting the numbers right.

A Little More Polish

After the workshop, I spent a couple more hours tightening things up and adding a demo mode so I could share the dashboard publicly without exposing private data. Demo mode swaps the data layer for realistic-but-fake numbers while keeping the real UI and interactions.

Why This Approach Works

Single source of truth: queries read from Supabase directly; no ETL.
Familiar login flow: no new accounts for the team to manage.
Fast iteration: Lovable handled most of the scaffolding so I could spend time on metrics.

Big thanks to Build First for the push to just ship it!

I Built an AI Stretching Coach for a Kids' Speaker (Yoto)

2025-10-10T00:00:00Z

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.

Matt Hamilton's blog

Adding an Analytics Dashboard for Toy Boat (in an hour)

The Setup

What I Built During the Workshop

A Little More Polish

Why This Approach Works

I Built an AI Stretching Coach for a Kids' Speaker (Yoto)

The Setup

The Tricky Part

What's Next