In the early stretch of caring for our baby, one of the most annoying problems was also one of the smallest: keeping track of feedings and diaper changes when more than two people were involved.

My spouse and I were not the only caregivers. Grandparents were helping too. That meant the same questions kept coming up in slightly different forms. Did the baby already eat? How much? Was that diaper logged or just mentioned out loud? The old post starts there, which still feels right to me, because that was the real design problem: not “build something with Alexa,” but “make shared care information easier to trust.”

Voice made sense because the context was physical. People had full hands. They were moving quickly. They did not need another interface to open, interpret, and remember to update. They needed to say something once and trust it would be captured correctly. The earlier version points to exactly that: voice as a hands-free way to log updates, with shared access across caregivers.

I did not choose Alexa because it was exciting. I chose it because paper notes were messy, texts were scattered, and the available apps either felt unproven or not worth a monthly fee for what we needed. I also wanted the logs to connect to a shared Notion setup my partner and I were already using. So the actual system became Alexa on the input side, Notion on the visibility side, and backend glue in the middle. The original post describes that stack pretty plainly: Alexa Skills Kit, Node.js, Google Cloud Functions at first, then AWS Lambda later to reduce latency, plus Notion integration for the shared log.

That migration to Lambda was not a flourish. It was a response to platform limits.

One of the harder constraints was Alexa’s timeout window. The old post cites an 8-second limit, which meant I could not be casual about database calls or multi-step logic. The backend had to stay lean enough to answer quickly.

The more human challenge was language. People do not all log the same event the same way. “Log a bottle.” “Record a feeding.” “A feeding from earlier today.” “A bottle at noon yesterday.” Getting the system to handle those variations well took more work than the basic happy path. The original version calls out that exact problem around phrasing past feedings and fine-tuning intents and slot values.

That was the useful part of the project for me. It made something obvious feel technical again: shared routines break down when the system around them is vague.

I also like this project because it stays small. It does not need to become a universal parenting platform to be worth talking about. It was a family coordination tool. It used familiar systems. It reduced one kind of friction at a time.

That feels like a good standard for design work in general.

Not every worthwhile thing needs scale. Sometimes the job is to make one recurring household question easier to answer, especially when people are tired.

That is what this project did. And that was enough.