From-scratch deep learning laboratory
Train neural networks.
Understand every gradient.
NeuraForge is an interactive machine-learning laboratory where you build, train, and dissect neural networks on classic benchmarks — spirals, circles, XOR — with every epoch of gradient descent visible in real time.
Design the architecture
Feel the gradients flow
Tune to convergence
Log the experiment
From-scratch engine
Backpropagation, Adam, momentum, and L1/L2 regularization — implemented in pure TypeScript. No TensorFlow, no PyTorch, no shortcuts.
Real-time visualization
Watch the decision surface bend, weights re-wire, and loss curves converge live — every frame is real gradient descent.
Experiment tracking
Every run is persisted with full hyperparameters and metrics. Compare architectures like a proper research log.
AI mentor
Professor Ada reads your loss curves and suggests the next experiment — like a teaching assistant who never sleeps.