> an in-house model we've been developing since 2019 We will be very interested in how you tackled the problem of "understanding" the input text - to different

> an in-house model we've been developing since 2019 We will be very interested in how you tackled the problem of "understanding" the input text - to differentiate it from current "fakeries of actual speakers". I.e. how you implemented the "intelligent" parts - or, which simulation of actual intelligent processing (if any) it contains.