In order for AI applications to be able to make the best possible use of the computational resources in which they will be executed, it is necessary to implement performance studies, through profilers, to locate and characterize the bottlenecks in their execution, followed by a search for solutions that may come ranging from restructuring the algorithm to the development of tools that enable better use of algorithms on the respective hardware platforms. Especially for AI, for example, it is important in this context to seek to be familiar with the latest mixed-precision processing technologies (double, single and half-precision), whether with regard to computational architectures, as well as the algorithms that best exploit the potential of this paradigm.