This repository implements a multi-head classifier using SetFit for query preprocessing, classifying queries into intent, domain, and human-in-the-loop (hitl) categories. Still need work..
It is a preprocessing step, I don't feel the need for creating separate classifiers. So you have shared embeddings and multiple heads for each task which i think is efficient.
I think for customer centric applications, analysing the query can be really crucial: decide where to route and what actions need to be taken, is the query relevant enough...
I am always open to interesting projects/opportunities and connecting with like minded people.