https://cs.stanford.edu/people/karpathy/convnetjs/
But even with libraries such as D3.js for visualization. Or Tensorflow.js for WebGL accelerated learning. Many scientists simply lack the front end skills necessary to create usable demos. Which can mean an opportunity to reach out to a team if you are looking for projects to work on.
https://bost.ocks.org/mike/algorithms/
We can also see where this is ultimately headed with the rise of electronic lab "notebooks" such as JupyterHub for data science. We want an intelligent interface, that could even understand human language to design experiments. It needs to be scalable and distributable to large enterprise research teams spanning multiple locations. Results have to be repeatable, obviously. And with a single click, publishable to the entire scientific community and general public.
What the future looks like is science "as a service"