alexcybernetic/playground-ai

playground-ai is a small collection of interactive learning pages that explain how neural networks work, one concept per page. Each lab is a single self-contained HTML file with no build tools and no JavaScript dependencies, so it can be opened directly in a browser from a downloaded copy or viewed online without installing anything. The pages use Google Fonts when a network connection is available, but fall back to normal system fonts and still work fully offline. The first lab, called Neuron Lab, shows what a single artificial neuron actually computes. Users can switch between a perceptron, which fires with a hard yes or no answer, and a sigmoid neuron, which produces a smooth graded output between zero and one. Sliders let you adjust each input value and its weight, plus a bias term, and an animated diagram shows every number flowing through the calculation alongside the full equation with real numbers filled in. Preset scenarios, like a loan approval or spam filter example, apply the same underlying math to different everyday situations. The second lab, Gradient Descent Lab, demonstrates how that same kind of neuron learns from data over time, using a training method called stochastic gradient descent to separate two overlapping groups of one hundred fixed points. It shows four connected views at once: a scatter plot with a moving decision boundary, a diagram of the current weights, the formulas for each training step with numbers substituted in, and a chart of the error decreasing over time. Users can play the training process at different speeds, step through it one point at a time, or jump ahead an epoch, and every run produces the exact same result because the starting conditions are fixed rather than random. Both labs share a consistent light and dark theme and are intentionally simple, deliberately leaving out more advanced training techniques so the underlying math stays visible rather than hidden behind shortcuts. The project is released under the MIT license, meaning anyone can use, copy, and adapt it freely, and one dataset is adapted from a Udacity course exercise on neural networks.