Advances in therapies like CRISPR editing or gene replacement offer the promise to directly ameliorate a rapidly growing list of human genetic diseases. However, robust cell-type targeted therapeutic delivery is lacking for these approaches, limiting their efficacy. In complex tissues like the brain, specific subsets of cells drive disease but are largely untargetable. Capsid-engineered viral vector approaches poorly target clinically actionable neurons or immune cells, while over-target the liver with lethal complications. We developed machine learning platform to write synthetic, fit-for-purpose cis-regulatory elements that can drive exquisitely precise gene expression, overcome the current barrier in gene therapies.