In this contribution we present a method to improve the accuracy of stellar measurements such as temperature, surface gravity, and metallicity, with a pure machine learning method. Our approach make use of measurements coming from large photometric sky surveys like SDSS and SMSS. This technique builds upon our earlier work, in which we standardized and refined spectroscopic measurements from various spectroscopic surveys such as APOGEE, GALAH, and LAMOST into a unified database. The latter was used to train an artificial neural network. By applying this method, we can calculate stellar properties with accuracy comparable to spectroscopic observations for a vast number of stars that have only been studied through photometry and with low resolution spectroscopy. Our measurements achieve precision levels of approximately 50 K for temperature, 0.08 dex for surface gravity, and 0.08 dex for metallicity. In particular, our approach remains reliable even when analyzing metal-poor stars, a category that has traditionally posed significant challenges for accurate measurement.