Buildings’ operational and embodied environments are responsible for a large portion of global greenhouse gas emissions. To reduce embodied carbon, eco-friendly materials and structural optimization are crucial. From a sustainability point of view, timber is considered one of the preferred material. Currently, mass timber building construction is experiencing a surge in North America and Europe. However, mass timber systems are generally more costly, compared to conventional systems built with other common structural materials. Structural design optimization can help in this regard and promote the construction of this sustainable alternative, to traditional systems. The current research is developing a method of optimizing glue-laminated timber (glulam) beam design through the use of reinforcement learning (RL). RL is a branch of machine learning where the agent interacts with an environment and learns through trial and error. Herein, the agent will take actions by picking glulam beams with a defined range of sizes, species and grades, and conducting the necessary design checks as per Canadian timber design standard, CSA O86-19. For the action taken, the agent receives rewards. The ultimate objective of the agent is to maximize the reward by minimizing the material cost. An RL agent, called the Proximal Policy Optimization algorithm, is trained to design the beam. It is found that the agent successfully designs the beam and minimizes the cost for a given beam length and loading condition.