The most common metric for CEST analysis is the magnetization-transfer-ratio asymmetry. Although qualitatively useful, it is affected by a mixed contribution from several exchange properties and requires experiment-specific protocol optimization. Herein, we propose a machine-learning framework for simultaneously tackling two challenging tasks: (1) automatic design of the optimal CEST acquisition schedule; (2) automatic extraction of fully quantitative CEST maps from the acquired data. The method was evaluated in simulations and phantoms at 4.7T. The resulting data acquisition and reconstruction times were 52 s and 36 ms respectively, providing quantitative exchange-rate and volume fraction maps with good agreement to ground-truth.