ANDREA ORTINO
AI DESIGN 1
Designing a Scalable and Modular AI Framework
PROJECT - TOWN OF ZOZ
ROLE - DESIGN LEAD
When I joined Studio Pixanoh, I was tasked to design an AI framework for enemies that could be reused across the entire roster, including bosses. The challenge was to design a flexible system adaptable to both gameplay and narrative needs.
We expanded Unreal’s behaviour-tree system by implementing modular tasks and plugins based on goals, triggers, priorities, and cooldowns. The framework supported both individual and group behaviours, with additional “phase” functionality for boss transitions and encounter escalation.
Once stable, I focused on documentation and team onboarding—recording walkthroughs, preparing annotated examples, and leading feedback sessions to ensure consistency. As a result, designers could create new enemy variations with minimal engineering support, accelerating production and maintaining design cohesion across
the roster.
IN A NUTSHELL
CHALLENGES
As a story-rich and combat-heavy action RPG, Town of Zoz required every enemy and boss encounter to reflect the game’s core narrative themes. Most of Town of Zoz’s enemies were inspired by food, with both their visuals and loot tied to real ingredients. For example, the “Avogato” featured a collar shaped like an avocado half and rewarded players with the same item upon defeat.
Because ingredients, farming, and cooking were central pillars of the experience, it was clear the game would require a large and varied enemy roster. For a mid-sized indie studio, this posed a major production risk: maintaining unique behaviours for every enemy would quickly exceed available resources.
The design challenge was to support visual and gameplay diversity with a system flexible enough to reuse and recombine core AI components across the entire roster.
APPROACH
RESEARCH
I began by researching existing AI paradigms, ranging from finite-state machines to utility- and goal-oriented systems, to determine which one best suited our scope. Working closely with the Tech Lead, we defined the feasible architecture within Unreal Engine and identified integration points for new behaviour-tree extensions.
To frame the design, I focused on five key behavioural questions:
-
How do enemies spawn?
-
What are their states?
-
How and when do they aggro the player?
-
How do they lose aggro?
-
What do they want to achieve while aggroed, and how?
IDEATION
Using one of the first enemy concepts provided by the Art Director, I established early design principles and core behaviour notes in Miro:
Concept of the Vermin Rat
Some quick ideas of behaviours and roles
The Vermin Rat, one of the first enemies players encountered, was a weak melee foe but dangerous in groups. Its ambush-oriented concept inspired one of the game’s core spawn behaviours: the Ambush Trigger, later generalised across multiple enemy types to create surprise encounters and pacing variety.
This early exploration led to several paper designs outlining spawn and aggro rules, followed by a behaviour-tree mock-up.
%20(2).jpg)
Aggro and spawn rules sketch
Rat Vermin AI paper design
MENTORING
Once the core system was stable, I led onboarding sessions for junior and mid-level designers, supported by walkthrough videos and annotated examples. These sessions established consistent implementation practices and enabled designers to create new enemy variations independently.
OUTCOMES
The modular AI architecture gave designers flexibility to create new enemy behaviours without additional engineering support. Over time, this framework scaled to more than sixty unique enemies and bosses, along with derivative variants. The Phases System allowed designers to script dynamic boss encounters that evolved during combat, introducing dramatic pacing shifts and moments of surprise.
READ MORE
You can read more case studies at the links below. For any questions or further details, you can send me an email at asortino11@gmail.com or a connection request on LinkedIn