No coding. No screens. No reading required. Just press, teach, and play.
The camera-powered AI toy. See it, snap it, train it.
Point the camera at two objects, a shoe vs a cup, a plant vs a toy. Press BTN 1 or BTN 2 to label each one.
Hit TRAIN. The device builds an image classifier in under 30 seconds, LEDs pulse as the AI learns.
Show new objects to the camera and watch it guess, then unlock gesture-powered games like Simon Says.
The microphone-powered AI toy. Hear it, record it, train it.
Make two sounds, a clap vs a whistle, a word vs a hum. Hold BTN 1 or BTN 2 to capture each sample.
Hit TRAIN. The device builds a sound classifier from your audio samples in seconds.
Make new sounds and test the AI, then unlock sound-powered games like Sound Detective.
Everything packed into one friendly little box.
MILTALK · ON-DEVICE ML
Captures what you show it for image classification tasks.
Listens for claps, whistles, words, any sound you teach.
Chunky, color-coded, for labeling, training, and testing.
Shows prompts, predictions, and training progress right on the device.
Miltalk isn't a simulation. Here's the real ML pipeline running on-device.
A lightweight pre-trained model (like MobileNet) converts each camera frame or audio clip into numbers that describe what's in it.
A fresh classifier is trained on top, just for your kid's two categories. This is called transfer learning, and it's how real AI engineers build models fast.
When kids press TEST, the model runs live, predictions come back in under 2 seconds right on the LED screen.
The freshly trained model becomes the controller for unlockable games, turning "training" into "setup for play".