In this research-driven project, I developed a system to automate enemy guard placement and coordination in video games, aimed at improving level design efficiency and gameplay balance. The goal was to help designers quickly prototype and adjust stealth or action levels without needing to manually place every guard or test countless patrol configurations by hand.

The system automatically analyzes a level’s layout and determines optimal guard positions and patrol routes, ensuring that each area is covered in a balanced and believable way. To help designers understand and fine-tune how secure a level feels, I introduced a pheromone-based simulation system inspired by swarm intelligence and natural pathfinding behaviors.

In this system, every guard “emits pheromones” along their patrol routes — invisible markers that represent the intensity of surveillance in a given area. As guards move and their vision cones overlap, these pheromones accumulate, forming a heatmap of perceived danger across the level. Red zones represent heavily guarded or frequently observed areas, while green zones highlight blind spots and potential infiltration routes. This gives designers an intuitive, visual way to evaluate how safe or dangerous different areas feel for players, without needing to manually playtest each configuration.

This approach merges AI behavior modeling and level design visualization, helping designers see the invisible logic of the game world — where enemies perceive danger, how players might approach, and where tension naturally builds. By turning abstract AI data into tangible visual feedback, the pheromone system bridges the gap between technical simulation and creative design intuition, making stealth and action levels faster to design, easier to balance, and more satisfying to play.

To complement the Guard system, I designed a Squad Manager system that enables groups of guards to coordinate tactics dynamically. Instead of behaving as isolated AIs, guards can now share information, assign roles, and react as a team — for example, flanking or reinforcing each other based on the situation. This coordination framework is powered by a lightweight, Utility AI–inspired role assignment system that evaluates each unit’s position and abilities to select the most effective group behavior in real time.