Abstract The transformative breakthrough of Google DeepMind’s AlphaFold2 on the reliability of sequence to protein structure prediction, demonstrated the power of machine learning approaches in advancing the study and engineering of proteins. Currently a number of inverse protein folding neural network models employ different objective functions in the design of proteins which come with trade-offs and can lead to adversarial sequence predictions. This project seeks to apply a different objective function to overcome limitations of current inverse protein folding models with the specific goal of predicting mutations that will increase the stability of therapeutic and diagnostic proteins. Additionally, AI- and Physics-based Simulation filters are integrated to enable the prediction of sequences that increase stability and retain function. It is hypothesized that by combining these AI tools with the experimental cell-free protein synthesis and stability/activity assays, rapid design-build-test-learn cycles can be performed to create AI models tuned specifically for the target protein. This technology is directly applied to the highly sensitive diagnostic reporter protein NanoLuc and the promising cancer therapeutic Onconase to expand their utility through enhanced stability.