We developed a Physics-Informed Machine Learning (PIML) approach to solve inverse problems. Traditional Multi-Disciplinary Design Optimization (MDO) require a large number of iterations to optimize a system for a given objective function. In this paper, we demonstrated the use of Deep Operator Networks (DeepONet) to solve an inverse MDO for a novel air-conditioning (AC) system that uses liquid desiccant technology developed by a company called Blue Frontier. Our PIML model takes velocity profiles inside the channels for a few designed cases to determine the optimum geometric parameters that satisfy the air conditioning constraints, boundary conditions and objective functions. The geometry of this air-conditioning system was composed of several channel pairs. Theses channel pairs were divided into the process channel to take the supply flow and the exhaust channel to return the flow. In both the exhaust channel and the process channel there were several geometric features that guided the internal flow to be distributed evenly throughout the channel. The size and the location of these features played a significant role to archive a uniform flow conditions and therefore they were the design parameters for this optimization problem. To optimize the parameters of geometric features in the CAD design, we took a three step process: a forward stage, a training stage and an inverse stage. The inverse stage could also be called testing stage or recalling stage. In the forward stage we first solved the Reynolds-averaged Navier–Stokes equations by finite-volume-method in the fluid domain using commercial off-the-shelf (COTS) computational fluid dynamics (CFD) tools. Then in the training stage, we built our physics informed deep operator network (PI-DeepONet) framework to define a functional relationship between the velocity profile throughout the fluid domain and the geometric variables. Here, the loss function, enforced all constraints and objective for the fluid domain. At the end for the inverse stage, we provided the desired and ideal velocity profile as an input for the model to get the optimum solution for the geometry.