Abstract: Strong gravitational lensing is a powerful probe that provides high-resolution views of the early universe and valuable insights into the nature of dark matter and dark energy. In this talk, I will present my analysis of high-redshift, lensed dusty star-forming galaxies observed with Atacama Large Millimetre/submillimetre array (ALMA). These early galaxies offer valuable insights into early galaxy evolution. I will discuss detections of molecular water and carbon monoxide emission lines, used to infer key physical properties of these galaxies. To recover intrinsic source properties, I will introduce an image-plane lens-modeling framework tailored for interferometric data from radio and submillimeter telescopes. The second part of my talk will focus on machine learning applications for cosmology and telescope operations. I will present simulation-based inference with neural ratio estimators to constrain cosmological parameters from lensing images, essential for analyzing thousands of systems expected from surveys such as Rubin LSST. I will also highlight machine learning models that identify and filter defective images to ensure reliable Rubin data processing. Together, these studies demonstrate how strong lensing and machine learning can jointly advance our understanding of galaxy formation and the dark universe.