Transaminases have been increasingly utilized as efficient biocatalysts in the synthesis of pharmaceutical intermediates, but a major drawback is their poor substrate acceptance, especially the limitation for the synthesis of sterically hindered chiral amines. Herein we report the engineering of a transaminase that can convert the ketone (6S,9R)-6-(2,3-difluorophenyl)-9-hydroxy-6,7,8,9-tetrahydro-5H-cyclohepta[b]pyridin-5-one to (5S,6S,9R)-5-amino-6-(2,3-difluorophenyl)-6,7,8,9-tetrahydro-5H-cyclohepta[b]pyridin-9-ol, a key intermediate for the synthesis of rimegepant, a CGRP antagonist for the treatment of migraine. Starting from an enzyme backbone with no detectable activity toward the desired ketone, a rational design approach enabled us to produce an enzyme variant with detectable trace activity. Then, by following various evolution strategies, including iterative saturation mutagenesis focused on a key loop and random mutagenesis of the whole sequence, further improvement of the activity was achieved. The resultant variant showed 99.0% conversion and >99.5% de for the desired reaction at the gram scale as well as at the kilogram scale to afford the product (5S,6S,9R)-5-amino-6-(2,3-difluorophenyl)-6,7,8,9-tetrahydro-5H-cyclohepta[b]pyridin-9-ol in 80.2% yield with 99.9% HPLC purity, thus showcasing promising potential for industrial application.