Saemil Engine
v1.0 / NumPy
- Vision: Checking…
- Names: Checking…
- Tabular: Checking…
NEW SESSION
What would you like to try?
Three small models, built on the same NumPy engine. Choose a task to start.
Predictions work offline. Explanations use Offline Template by default.
VISION / 01
Draw a digit. See what it sees.
Use a mouse, pen, or touch. Prediction runs when you finish a stroke.
28 × 28 input / light ink on dark
Keyboard alternative: loads a labelled MNIST test digit and classifies it.
MODEL OUTPUT
MNIST / 784 → 128 → 10
101,770 trainable parameters
NAMES / 02
A few letters. A new possibility.
A character model samples name ideas. A fixed seed repeats the same list.
CANDIDATES
TABULAR / 03
Explore a completed shopping session.
Change a value to update the estimate. The model uses all six features below.
ESTIMATED PURCHASE PROBABILITY
Initial values and slider limits come from the loaded model's training data.
SESSION / THREE HEADS
One engine. Three different answers.
Each head predicts independently. The explanation brings the results together.
AI BRIDGE / EXPLANATION MODE
Choose how to explain the results.
The three NumPy heads make the predictions. This setting only changes the prose.
If a provider fails, Saemil shows the offline explanation and its fallback reason.
A local CLI never silently switches to a paid provider.