Thermoelectric generators offer a promising approach for harvesting waste heat from both natural and human-made sources, enabling sustainable electricity generation. While geometric design plays a crucial role in optimizing device performance, conventional approaches remain confined to simple configurations, limiting efficiency improvements. This constraint arises from the complex interplay of multiphysical interactions and diverse thermal environments, which complicates structural optimization. Here, we introduce a universal design framework that integrates topology optimization (TO) with additive manufacturing to systematically derive high-efficiency thermoelectric 3D architectures. By formulating an optimization problem to maximize power generation efficiency, our approach explores an unprecedentedly large design space, optimizing the geometries of thermoelectric materials across diverse thermal boundary conditions and material properties. The resulting TO-derived geometries consistently outperform conventional cuboids, demonstrating significant efficiency gains. Beyond in-silico studies, we provide theoretical insights and experimental validation, confirming the feasibility of our design approach. Our study offers a transformative way for enhancing thermoelectric power generation, with broad implications for next-generation sustainable energy technologies.