The Lab
Interactive tools for hands-on machine learning
Experimental software, algorithmic demonstrations, and proof-of-concept models you can adjust and run directly in the browser. Every tool exposes its assumptions, so you can see how the parameters change the outcome.
Interactive tools
Adjust the parameters, watch the model respond
Demonstration 01
Gradient descent visualizer
Set the learning rate, initialization, and loss surface, then step through each update. Sliders control the parameters while the trajectory redraws in real time, making convergence and divergence easy to compare side by side.
Demonstration 02
Decision boundary explorer
Tune the regularization strength and model complexity to see how a classifier's boundary shifts across a two-dimensional dataset. Visual outputs update live, with source code linked for every configuration.
Prototype 03
Attention architecture sandbox
A proof-of-concept implementation for inspecting attention weights across token sequences. Still in active development, it shows the trade-offs between sequence length, memory use, and interpretability, with the code open for review.