# Blog · 1 June 2026 · 1 min read
A smart speaker in two weekends
Ten years ago this took a whole team.
It is amazing how far technology has come. Ten years ago you needed a whole team to make a smart speaker.
Now one person and two weekends are enough.
Inside there is an ESP32-S3 with an XVF3800, and a local wake word. It connects to a backend in my home Kubernetes through STUNner.
Text-to-speech and back is done by ElevenLabs.
For brains: GPT-5.4, with Hermes as the harness.
Obviously it still needs tools, such as a timer, and a way for the agent to send something to the speaker on its own initiative.
But it is cool, right?! What a time to be alive...