Inspect the learning process
Watch decision regions, errors, confidence, loss, weights, and layer flow update while the model learns.
01 — Decision Boundary
Watch the
model think
in real time.
The decision boundary canvas updates live with every gradient step. Switch between Boundary, Errors, and Gradients view modes — or click any point to inspect its full forward pass, layer activations, and confidence score.
Confidence zonesMisclassification lensForward-pass inspector
02 — Live Metrics
Every number,
explained.
Track train and val loss curves in real time. Spot the moment your model starts overfitting. Compare generalization gaps across architectures and learning rates without writing a single line of code.
03 — Internal State
Open the
black box.
For real.
The weight heatmap shows every learned parameter — positive weights in violet, negative in pink. Pair that with the gradient flow view to see which layers are learning and which are stuck.
✦ Explain Mode
→ Model is likely underfitting — try more neurons.
→ Learning rate may be too high — loss is unstable.
→ Generalization gap is small — model is healthy.
Dataset Preview · 320 rows valid
| x | y | label |
|---|---|---|
| 0.12 | 0.88 | 0 |
| -0.40 | 0.20 | 1 |
| 0.60 | -0.10 | 0 |
| -0.22 | -0.71 | 1 |
| 0.91 | 0.34 | 1 |
| ... | ... | ... |
✓ Auto column detect✓ Numeric validation✓ Normalize to [−1, 1]✓ 320 valid rows
04 — Custom Data
Your data.
Your model.
Your rules.
Drop any CSV with two numeric columns and a binary label. Axon detects columns, validates rows, normalizes values and trains on your data instantly — no backend, no upload, no waiting.
Drag & dropClick to uploadSample CSVManual column select
Bring your own CSV
Import a 2D binary dataset, auto-detect columns, validate rows, normalize values, and train locally.
Compare experiments
Save runs, export summaries, download JSON, and compare architecture choices across datasets.