Origin

I wanted a project where I could use machine learning and natural language processing. While I was mulling over project ideas and watching Westworld, a friend sent me a link to the WoW Classic announcement. That gave me the idea for Westworld of Warcraft.

What I wanted Westworld of Warcraft to be was a private server where there were other “players” that would be involved with the gameplay. Back in 2005, I remember playing with a number of memorable players each with personality quirks; an experience I hope to emulate with different hosts.

Goals

So here are the goals of the project (I will update them as I go):

~~Create programs that can interact with the game and consume generated data from the game.~~ (Completed)
~~Funnel consumed data to instances into Microsoft Azure.~~ (Completed)
Build machine learning models around the consumed data to mimic players playing the game. Including:
- Questing
- PVP
- Raiding
- Grouping
- Being Apart of the Economy
Build natural language processing models to:
- Allow bots to converse with players
- Spawn bots in response to continuing chat requests (LFG,LFM,WTS,etc.)
~~Change the bots to no longer need the client and use the server connection directly~~ (Completed)

Where those goals ended up: bots now run in two interchangeable modes - injected into a real game client for parity, or fully headless against a C# implementation of the game protocol. Behavior is structured as a four-layer hierarchy, objective sequences are generated at runtime from the world database rather than scripted, and every character carries a personality derived deterministically from its account name. Machine learning and the persona dialogue runtime are both advisory: they break ties and draft text, and neither can choose an action or touch the world.

Series

The architecture is covered end to end in a thirteen-post series: