Neural architecture search with Simmulated Annealing

im trying to build a neural architecture search that would use annealing to decide ALL the parameters such as the number of conv layers, filters, dense layers, optimiser used, dropout, max pooling etc