AI framework within a simulation game that integrates an LLM for decision-making and dialogue, supported by a graph database for tracking and contextualizing scenarios through psychological models
Team Members: Lawrence J. Klinkert
This project demonstrates a novel AI framework within a simulation game, where players raise a dragon whose personality, emotions, and goals evolve based on interactions. The game integrates an LLM for decision-making and dialogue, supported by a graph database for tracking memories, goals, and relationships, contextualizing scenarios through psychological models.
The dragon's behavior is dynamically shaped by systems including a memory module that shifts short-term experiences to long-term memories, and an emotional model based on the OCC framework, which interprets events in real time. These systems interact with a PAD-based mood model and a goal system that prioritizes the dragon’s needs by linking gameplay actions to outcomes.
Social relations and personality traits modify how the dragon experiences and recalls events, adding depth to its responses and decision-making. Using Retrieval-Augmented Generation (RAG) and knowledge-based queries, the LLM crafts responses that adapt to the dragon’s psychological state and social context, making this a scalable model applicable to educational simulations, adaptive AI, and character-driven storytelling.
