Despite the increasing capabilities of computers to master sophisticated human-like tasks and the recent explosive new wave of Machine Learning-based methods, the interior design field still remains a hard-to-master area, without robust, mature models that could compete with the expertise of humans in the field. This is an exciting new are for Artificial Intelligence in general, still in its very early stages of development, both in terms of models performance and in terms of specialized data availability. Most current applications of this type of models remain only in the area of virtual reality. Veering away from this trend, the current thesis proposes an end-to-end proof of concept for applying Machine Learning techniques to realistically asses the quality of professional and realistic room furniture layouts. We do so by proposing a learning-based scoring function comprising various interior design guidelines, ergonomics and plain common sense metrics. We further propose a stochastic optimization proof of concept based on Simulated Annealing techniques, aiming to generate new plausible and pleasant furniture layouts that obey the strict regulations of interior design. This proof of concept represents a first step towards the final goal of developing a software tool that would eventually demonstrate that real world, furniture layouts of professional quality can be obtained in an at least semiautomatic manner, using an energy function that analytically represents, as cost terms, various furniture functional and style interdependencies, common practices in relative furniture positioning in a room and other ergonomic factors that contribute to obtain a pleasant, livable room. Using machine learning to adapt the ranking function parameters across various types of rooms and sophisticated furniture objects, the method is supposed to scale in modeling complex interior design know-hows, hard to be modeled mathematically or learned directly by a purely data-oriented model.