Lower limb prostheses are designed to replace the functions and form of the missing biological anatomy. These functions are hypothesized to improve user outcome measures which are negatively affected by receiving an amputation – such as metabolic cost of transport, preferred walking speed, and perceived discomfort during walking. However, the effect of these design functions on the targeted outcome measures is highly variable, suggesting that these relationships are not fully understood. Biomechanics simulation and modeling tools are increasingly capable of analyzing the effects of a design on the resulting user gait. In this work, prothesis-aided gait is optimized in simulation to reduce both muscle effort and peak loads on the residual limb using a generalized prosthesis model. Compared to a traditional revolute powered ankle joint model, a two degree-of freedom generalized model reduced muscle activations by 50% and peak loads by 15%. Simulated prosthesis behaviors corresponding to the optimal gait patterns were translated into a two degree-of-freedom ankle-foot prosthesis design with powered bidirectional linear translation and plantarflexion. The prototype is capable of delivering up to 171 N-m of plantarflexion torque and 499 N of translation force, with 15° dorsi-/35° plantarflexion and 10 cm translation range of motion. The mass and height of the ankle-foot are 2.29 kg and 19.5 cm, respectively. The mass of the entire system including the wearable offboard system is 8.58 kg. This platform is designed to emulate the behavior of the simulated prosthesis, as well as be configurable to emulate alternate behaviors obtained from simulations with different optimization objectives. The prototype is controlled to replicate simulated walking patterns using a high level finite state controller, mid-level stiffness controller, and low level load controller. Closed loop load control has bandwidth of 15 Hz in translation and 7.2 Hz in flexion. Load tracking during walking with a single able-bodied human subject ranges from 93 to 159 N in translation and 4.6 to 21.3 N-m in flexion. The contribution of this work is to provide a framework for predictive simulation-based prosthesis design, evidence of its practical implementation, and the experimental tools to validate future predictive simulation studies.