NANIZHOUZ
A first person horror game about a house that does not want you to leave. Zombie AI with real sensory systems, limb damage, a vendor economy and a chainsaw boss who is genuinely a problem.
NANIZHOUZ is the biggest game project I have taken on, and the one that taught me most about scope.
The enemies
The zombies are built around senses rather than states. Each has sight with a real cone and occlusion, hearing with falloff, and a memory of where it last believed you were. It does not know where you are. It knows where it thinks you are, and it goes there.
That single design choice produces most of the good moments. Throwing something to move a zombie’s belief works because belief is a thing the system actually models. Hiding works because the enemy commits to a stale hypothesis for a while before giving up.
Limb damage is per-part with dismemberment, which changes behaviour rather than just visuals. A zombie missing a leg is slower and still coming.
The house
Puzzles are physical rather than inventory-based: a television, a dollhouse that mirrors the building, a boiler that has to be brought back up. There is a vendor with an upgrade economy, a throwable system with fire that propagates across surfaces, and a darts minigame that exists purely because the room needed something to do in it.
The chainsaw boss remains the hardest thing in the game to balance. Every time I make him fair he stops being frightening.
What I learned
The project has a full design document, a marketing plan and a launch checklist, and writing those was more useful than any of the code. Roughly four hundred gigabytes of assets is a decision that constrains everything downstream, and I would make a different one now.
It is presented here as a case study with footage rather than a playable build. Streaming that asset volume to a browser is not a realistic proposition, and pretending otherwise would waste your time and mine.