This document describes a Bayesian adaptive dose selection procedure using semi-parametric dose-response modeling. It involves modeling dose-response relationships non-parametrically using a monotonicity constraint, applying Bayesian model averaging over different dose-response shapes, and performing predictive probability calculations using importance sampling to determine the dose to take forward at an interim analysis based on predictive probabilities of success criteria being met. Simulation results demonstrate the procedure can correctly identify the optimal dose and uses fewer patients than a non-adaptive design.