Energy dissaggregation - NILM - anyone tried to figure this out for themselves?

I have recently become interested in the topic of non-intrusive load monitoring.
From English NILM or energy dissaggregation These are some of the articles on this topic:
https://towardsdatascience.com/energy-managem...me-non-intrusive-load-monitoring-3c9b0b4c8291
https://en.wikipedia.org/wiki/Nonintrusive_load_monitoring
In a nutshell, the idea is to detect by the load characteristics what specific device is running and draws current for the input data having a total summed total load.
I found some projects on Github but I don't really understand what needs to be done.
I know that I need to have machine-trained datasets that will be able to extract from the characteristic spikes etc from the summed current or power graph that the device is running. Looking at my own graphs many times I know what is working at the moment.
There was once a topic that something similar was done by someone here on the electrode, but I can't find it now.
BTW, I know that I can connect a smart socket to any device and have everything obvious. That's not what I mean.