Modeling urban microclimate accurately is challenging due to the high surface heterogeneity of urban land cover and the vertical structure of street morphology. Recent years have witnessed significant efforts in numerical modeling and data collection of the urban environment. Nonetheless, it is difficult for the physical-based models to fully utilize the high-resolution data under the constraints of computing resources. The advancement in machine learning techniques offers the computational strength to handle the massive volume of data. In this study, we proposed a machine learning approach to estimate point-scale street-level air temperature from the urban-resolving mesoscale climate model and a suite of hyper-resolution urban informatics, including three-dimensional urban morphology, parcel-level land use inventory, and a dense weather observation network. We implemented this approach in the City of Chicago as a case study. The proposed approach vastly improves the resolution of temperature predictions in cities, which will help the city with walkability, drivability, and heat-related behavioral studies. Moreover, we tested the model's reliability on out-of-sample locations to investigate the application potentials to the other areas. This study also aims to gain insights into next-gen urban climate modeling and guide city observation efforts to build the strength for the holistic understanding of urban microclimate dynamics.