AI NPC Automation & Behavior Development

I’m building a proof-of-concept where an autonomous agent can watch a virtual world through an SDK, reason about what it sees and then act just like a fully fledged non-player character. The environment itself is NPC-centric, so everything you create has to plug cleanly into the existing observation / action interface I already expose.

What the agent must ultimately demonstrate is the full spectrum of believable behaviour:

• NPC behaviours to cover: combat actions, rich social interactions and day-to-day resource management.
• Within combat the agent needs strategic movement, solid weapon-handling logic and the ability to switch into defence tactics when the situation flips.

I’ll provide the SDK (Python and C++ bindings) plus stubbed observation data. Your job is to wire up the decision-making layer—whether you favour behaviour trees, GOAP, deep reinforcement learning or a hybrid approach is up to you—as long as the final agent can:

1. Parse state frames coming from the SDK in real time.
2. Select an action that makes sense for the current goal.
3. Return that action through the SDK with