Timosgroup

se, ¿cuál? (“That one — which one?”) is video perception for lab demos. In my neuroscience lab in Bolivia, students learn assembly from recorded demos where the instructor says “connect this here” without naming the part. The system resolves that reference: Muse Voice Transcribe gives the phrase and its timestamp, Muse Spark 1.3 links the gesture to the phrase and picks the moment, SAM 3.1 segments and numbers every candidate part in that frame, Spark picks one, and SAM 3.1 tracks it through the clip. When nobody could tell, it abstains (“I can’t tell which part”) instead of guessing. On 10 real breadboard clips recorded on Nov 4 and run once with no tuning: 5 of 8 parts identified, 2 of 2 correct abstentions, p95 39.7 s, ≈ USD 0.03/clip. Every failure is published clip by clip. Everything was built during the hackathon; before Oct 26 only idea notes existed.