Shakeolin | Studio Side Project

THE QUESTION
Can a studio ritual run on a physical object, with no screen and no buttons?

THE CHALLENGE
Our studio picks a new host for a recurring team meeting each time. I used it as an excuse to learn what I don't do every day: integrating electronics, sensors, on-device AI, audio and physical behavior into one small object.

MY CONTRIBUTION
I built it end to end, in stages:
Breadboard first: an ESP32-S3 with an IMU accelerometer, audio amplifier and ePaper display, then a working mock-up that announced the next host in Teams through Power Automate.
Shakeolin and LilyPad: a frog you lift and shake to reveal the host, sitting on a LilyPad base. Form explored from 2D sketches to 3D, then detailed in SolidWorks with a flexible shaft mechanism and 3D printed.
Custom PCB: components selected with AI assistance, schematic and layout in KiCad, boards ordered and assembled, then packaged into the enclosure.
TinyML with Edge Impulse, on the device: shake-gesture recognition from the IMU, then offline YES/NO keyword spotting, with a captive-portal dashboard as fallback.

THE OUTCOME
A working object demoed live to the studio. The open question I'm still pushing: how far a €9 chip can take keyword-spotting accuracy beyond the ~80% it reaches today.