A complete local voice pipeline: she speaks in her own cloned voice, hears you through your microphone, and answers from a model running on your hardware.
She remembers you between power cycles, wakes to her name, and acts on rules and habits instead of waiting to be asked.
Her speech stops sounding read-aloud, she reacts to how you say things, and the project gains the debugging and measurement habits a growing system needs.
The pile of working scripts becomes one designed system with a startup that validates itself — and a message protocol to the microcontroller her body will use.
She survives her own failures, her reactions live in data, and real sensors and actuators join the loop under test.
A permission system between her mouth and your shell, hardware behind swappable drivers, and the whole stack running as a service you can hand to someone else.
The physical GLaDOS: printed parts, a moving eye with a light ring, a robotic arm, microphones and a speaker, all calibrated and tested as one machine.
Flash, verify and productionize a Jetson Orin Nano, migrate the whole stack onto it, and measure what the GPU buys her voice and her brain.
She searches her own knowledge, sees the room, starts speaking before the model finishes thinking, and takes over the smart home.
Demonstrate supervised arm reaches to manually supplied, calibrated coordinates, compare camera-relative depth estimates, measure vocal cues, and review a day of operation. Camera-to-arm targeting remains outside this build.