Proteins are biological macromolecules with a variety of functions that are necessary to sustain life. These functions are dependent on the inherent flexibility of proteins, that can be induced through molecular interactions to change conformations, transitioning between inactive and active states. This ability is vital for the survival of cells, as it allows protein switches to transmit biochemical information throughout the cell, linking extracellular sensing, intracellular signaling, and cellular response. The capability to recapitulate the inducible characteristics of natural protein switches in designed proteins is desirable, as it would allow for built-in regulatory mechanisms, with potential applications in controlling the activity of protein-based or cell-based therapies. However, despite tremendous advancements in computational protein design in recent years, most designed proteins are limited to a single state. Additionally, many existing tools do not incorporate non-protein components into the design and modelling process, making it difficult to design de novo protein switches that rely on such inputs. To overcome these limitations, this work focuses on designing protein switches with distinct inputs and outputs. First, we aimed to integrate regulatory mechanisms into interleukins. Recombinant interleukins represent a promising therapeutic for their ability to modulate immune responses. However, their clinical success has been limited by off-target effects and toxicities. To address this problem, interleukins were designed to incorporate inhibitory domains, such that they are inactive. These inhibitory domains could be released by the addition of a small molecule, therefore activating the interleukin. This approach was generalized to three different interleukins that all showed small molecule induced activation. Next, small molecules were utilized to induce protein interactions directly, effectively functioning as an ON switch. Previously in our lab, MaSIF was developed, a geometric deep learning framework that fingerprints the surface of proteins based on geometric and chemical features. By inverting these fingerprints, this method was shown to generate de novo site-specific protein binders. Here, we aimed to generalize MaSIF to fingerprint protein:small molecule target complexes and design de novo protein binders specific to the complex. Binders were validated against three protein:small molecule complexes and were shown to activate function in cell-based systems dependent on the presence of their respective small molecule. Finally, we aimed to mimic natural proteins by incorporating phosphorylation as a mechanism to drive conformational changes in de novo proteins. Four different folds were validated for phosphorylation dependent switching. Furthermore, phosphorylation specific binders were validated against one such protein. This phosphorylation dependent interaction was shown to regulate transcription in cell-based systems.