Machine Learning applications such as the hybrid-CNN (Guigluon et al. 2024) allows to homogeneously combine Gaia RVS spectra, photometry (G, BP, RP), parallaxes and the XP coefficients to obtain precise stellar parameters down to S/N=15. In this contribution, I will focus on the scientific applications and new results facilitated by this machine learning approach and the new dataset. Firstly, in Nepal et al. 2024, we investigated a homogeneous and extensive sample of super-metal-rich stars to establish tighter constraints on the epoch of bar formation and its impact on the Milky Way's star formation history. We find hints of recent bar activity at ~3-4 Gyrs. Secondly, in Nepal et al. (submitted), we explore the very early evolution of the Milky Way disc using a set of over 200,000 Main Sequence Turn-Off (MSTO) + Subgiant (SGB) stars with precise ages including 8500 stars with [Fe/H]<-1.0. We report MW thin disc formation stared already at first billion year after Big Bang. And, both the old thin and thick discs were splashed at around 9 to 10 Gyr. Finally, in Nepal et al. (in preparation), for the first time with Gaia-RVS data, we peer into the MW Bulge to study our Galaxy's ancient stellar populations.