For the uninitiated, it can be roughly seen as detection of a sequence of musical notes. Raga is a term for a particular scale of notes (both ascending and descending).
Until now, this has mostly been in the domain of research and there is a ton of published literature out there. At the very basic level, if you have just voice, it is trivial to apply a pitch detection algorithm like YIN to get a pitch estimate and then analyse the sequence to figure out the raga. This doesn't work as well in a concert setup where speeds are higher due to gamakas, different instruments are used alongside and counterpoint melodies may make the music polyphonic. A lot of papers apply a variety of ML models (neural nets and otherwise) using several different features (cepstrum and mel-cepstrum, pitch distributions etc) with varying results.
So this is an interesting exercise in Signal Processing and Machine Learning. If anyone else is working on or has worked on this, I'd love to hear from you.