COVID-19 has become a worldwide pandemic whose rapid spread and mortality rate threatens millions of lives and the global economic system. Developing effective treatment such as neutralization antibodies is an urgent need. We propose here to develop a new method to design antibodies strongly bind to the SARS-CoV-2 receptor binding domain (RBD) that is necessary for viral entrance to human cells. We will develop a novel approach that combines directed evolution, deep sequencing and interpretable neural network models to efficiently identify strong and specific antibodies. This method will allow analyzing large sequencing data sets of antibody variants against the SARS-CoV-2 RBD in order to derive superior binders that do not exist in the original library. Iteration through directed evolution and computational design will efficiently identify neutralization antibody candidates that can be used as potent therapeutics to treat COVID-19.