Deficiencies or abnormalities in protease activity are common to many diseases, therefore proteases present attractive targets for use as therapeutics. However, broader use of proteases in the clinic is limited by their substrate specificity, since off-target activity would lead to unwanted side-effects. To enable engineering of human protease activity and specificity, we developed a cell-based assay in yeast to screen protease libraries for cleavage of a co-expressed fluorogenic peptide substrate exhibiting Förster resonance energy transfer (FRET). Use of the FRET reporter substrate enabled screening large protease libraries using fluorescence activated cell sorting for the activity of interest. This method was applied to human kallikrein 7 (hK7) to identify variants that selectively cleave the central hydrophobic core of the amyloid beta (Aβ) peptide involved in the pathology of Alzheimer’s disease. hK7 cleaves Aβ in vitro within the central hydrophobic core KLVF↓F↓AED (Aβ8) thereby preventing aggregation. Given that hK7 exhibits modest activity towards Aβ, but prefers Tyr over Phe at P1, we sought to improve activity and narrow Aβ specificity towards substrates with P1 Phe. Screening randomly mutated hK7 libraries yielded a protease variant with substantially improved Aβ selectivity, 16-1000 fold improved resistance to inhibitors, and reduced toxicity to PC12 neuronal-like cells. This methodology thus provides a powerful high-throughput capability to evolve human proteases for specific degradation of proteins implicated in a disease state.