Eigen RadarAI
Analysis

Sean Hollister adds speech and readable replies to a homemade Muse wearable

Sean Hollister used Claude Code to modify a Muse-connected development kit after encountering slow responses, missing speech and hard-to-read replies. His personal wearable gained conversation history, spoken answers and remote camera controls. The experiment, published October 31, also included a virtual pet. Hollister has no programming experience and says he cannot judge the quality of the generated code.

Artificial Intelligence··Morning
Wide view of a sunlit home dining nook with a small wearable development kit, its screen facing away, a neck cord, face-down phone and USB cable on a wooden table.

A personal build begins with Meta’s sample code

Gadgets journalist Sean Hollister connected a Waveshare ESP32-S3 1.75C development kit to Muse, Meta’s cloud AI service, for a project published on October 31. The touchscreen board is among the hardware Meta recommends to developers. Hollister used Anthropic’s coding assistant Claude Code to install Meta’s software from GitHub, then paired the device with the Muse phone application. He carried out the experiment on a separate kit from Meta’s forthcoming Muse Charm.[1]

Custom changes make replies easier to use

At first, spoken requests were treated as recordings to transcribe, replies arrived slowly and the device stayed silent. Longer answers were difficult to follow on its small display. Hollister asked for a conversation log that could be swiped through and added spoken replies using ElevenLabs, a speech-generation service. Camera changes introduced remote shutter control and corrected an upside-down front-camera view. A lower-quality preview over Bluetooth was another addition.[1]

A virtual pet joins the modified wearable

Further modifications added avatars and a virtual pet needing food, rest, play and bathing. To address recordings cut short when the talk button was released early, the software was changed to send audio only when it exceeded one second. Hollister says the remaining response delays varied with Meta’s servers. He used a guest wireless network and, as someone without programming experience, says he cannot assess the generated code’s quality.[1]

References

  1. News sourceThe VergeThe Verge’s wearable Muse experiment exposes gaps in the starter software↩1↩2↩3