It’s a bit niche, but I’m really proud of it so I figure I’d show it here anyway. 
in biophysics, people frequently analyze two correlated fluorescence signals under a microscope, which are recorded as two time series. One of those signals comes from a fluorescent donor dye, the other from an acceptor dye. The distance-dependent transmission of energy between these two (Forster resonance energy transfer, or FRET for short) shows up as anticorrelation in their fluorescence signals. The app I made, FRETboard, allows you to train an algorithm by correcting classification of a few of these time series by hand. Right now it supports a few kinds of HMMs but it’s modular so that you may write different algorithms for it as well.
On the Bokeh side of things many of the GUI elements worked straight away, but I did have to figure out how to up- and download single and multiple files at once through the GUI and process keystrokes - which was an adventure given my lack of prior JS experience, haha! Code and install instructions are here: GitHub - cvdelannoy/FRETboard: Flexible semi-supervised classification for FRET traces