Sign in to use the AI Lab
Training happens on your device, but an account keeps your experiments organized.
Train tiny models,
right in your browser.
Three hands-on labs. Real neural networks, real training, your own data – no server, nothing uploaded. Pick a lab to start.
info Training runs on your device – small datasets train in seconds; phones handle the demo sizes best. Desktop recommended for bigger runs.
dataset Step 1 – Your examples
Add at least 4 examples across at least 2 labels. More varied examples train better models.
model_training Step 2 – Train
Dense network: 64 → dropout → 32 → softmax. Vocabulary capped at 500 words.
science Step 3 – Test & export
dataset Step 1 – Your image classes
Create at least 2 classes and add at least 5 images to each. More images train better models.
model_training Step 2 – Train
Frozen MobileNet turns each image into an embedding; we train a small dense head on top (128 → softmax). Backbone download happens once.
science Step 3 – Test & export
dataset Step 1 – Your CSV data
Paste CSV with a header row. Feature columns must be numeric – the target can be numeric (predict a number) or text (predict a category).
model_training Step 2 – Train
Dense network. Features are z-score normalized (stats saved for inference).