The recovery of low- to medium-temperature waste heat via Organic Rankine Cycles (ORCs) has become increasingly important for industrial and energy systems driven by sustainability demands. In compact ORC turboexpanders, high specific power and large pressure drops are often achieved using single- or two-stage turbines operating in transonic or supersonic regimes. Under these conditions, conventional design heuristics are challenged by complex compressible-flow physics and real-fluid effects. Supersonic axial turbine stator blades are typically designed using the method of characteristics (MOC), which is applied only to the supersonic (diverging) portion of the blade passage, while the converging region is prescribed heuristically [1]. This paper proposes a systematic inverse design approach based on physics-informed neural networks (PINNs) for the complete supersonic stator blade geometry of an ORC Laval turbine. The proposed PINN framework [2] embeds the governing partial differential equationsof irrotational compressible flow and incorporates real-gas thermodynamics of organic working fluids to ensure physical consistency. Training targets are prescribed along the mean flow line within the inter-blade channel. Blade surface coordinates are then reconstructed by deriving suction- and pressure-side contours from stream functions predicted by the network. The resulting blade geometries are validated using high-fidelity CFD simulations of inviscid flow, demonstrating close agreement between the mean-line flow distributions and the design specifications. Subsequent turbulent CFD simulations confirm that the PINN-designed supersonic stator blades support smooth, shock-free flow fields. Furthermore, a single-stage transonic ORC Laval turbine equipped with the designed stator blades achieves efficiency comparable to that of a multistage subsonic Rateau turbine operating under the same expansion conditions. Overall, the proposed method integrates advanced machine-learning techniques with physics-based modeling, providing a robust and verifiable framework for supersonic blade design that effectively bridges data-driven approaches and classical CFD practice.