The enemies run on Blueprint-based AI built from the ground up, no pre-made behavior packs. Each enemy uses an AI Controller paired with a Behavior Tree and Blackboard to manage decision-making. The Blackboard holds key variables like player location, attack range, and current state, while the Behavior Tree reads those values to decide what the enemy does next: patrol, chase, attack, or retreat.

For movement, I set up custom waypoint navigation using Nav Mesh combined with a series of target points placed throughout the level. Enemies patrol between these points until a detection trigger, usually a combination of line trace checks and a proximity radius, spots the player and switches their state from patrol to pursuit.

Attack behavior runs through custom event graphs rather than default combat templates. Each enemy type has its own attack function bound to animation notifies, so damage only triggers at the correct frame of the attack animation instead of the moment the animation starts. I used timers and cooldown variables to prevent enemies from spamming attacks, and added randomized delay ranges so groups of enemies don't all attack in sync.

Building it this way meant every piece, detection, pathing, attack timing, state switching, had to be wired manually node by node. It took longer than plugging in a template, but it meant I could shape exactly how aggressive, how smart, and how fair each enemy felt.