The Least Squares Kinetic Upwind Method (LSKUM) is a meshfree scheme that belongs to the class of kinetic schemes for compressible flows. The core idea of LSKUM is to introduce upwinding into the governing equations through kinetic flux vector splitting, followed by the approximation of spatial derivatives using a weighted least-squares formulation. Over the past two decades, LSKUM has been successfully applied to a wide range of fluid flow problems. Central to the robustness and accuracy of the LSKUM solver is the goodness of the connectivity, which can be quantified by the condition number of the weighted least-squares matrix. In computational domains with highly stretched or anisotropic connectivity sets, these matrices often become ill-conditioned, resulting in a loss of accuracy, oscillatory convergence behaviour, or even code divergence. Although various strategies have been proposed to overcome these challenges, they are heuristic in nature and do not aim to minimise the condition numbers. This paper presents the development of optimally weighted LSKUM that yields minimal condition numbers. The optimal distribution of weights is found using a gradient algorithm combined with discrete adjoints. The robustness and accuracy of the optimally weighted LSKUM are assessed by applying it to the test case of inviscid subsonic flow over the MDA three-element high-lift configuration on a highly stretched point cloud. Numerical results have shown that the optimally weighted LSKUM yields a more accurate lift with higher suction peaks in the surface pressure distribution and better residue fall compared to the other choice of weights. In the full paper, the basic theory of optimally weighted LSKUM and detailed numerical results will be presented.