LLM-Driven RPG Game

This project investigates how large language models can be integrated into game systems beyond dialogue generation alone. Existing AI native games often use LLMs for storytelling, conversational play, or isolated language-driven mechanics, leaving a gap between narrative interaction and structured gameplay execution. To address this, the project developed a small 2D AI native role-playing game prototype in Unity, where players interact with LLM-driven non-player characters using natural language. The system uses Gemini through an API, with prompt constraints and structured JSON outputs used to translate player input into dialogue, quests, objectives, and rewards. Unity then parses these outputs and executes them through deterministic gameplay systems, allowing the LLM to influence the main quest loop without directly controlling game state.

The full write-up — architecture, the memory system, and how the quest generation pipeline works — is still in progress. Check back soon.