GameMind Agent is a conversational AI platform I developed at Schoene Neue Kinder GmbH for game developers. It uses LangGraph to orchestrate multi-step conversations and PostgreSQL with vector embeddings for semantic memory, so users can recall and refine game ideas across sessions.
What I built
- Semantic memory: Stored and retrieved game-design concepts via embeddings so the agent could reference past ideas accurately.
- Automated visualization: Turned high-level ideas into structured outputs that supported quick iteration.
- LangGraph flows: Designed the graph so the agent could switch between recall, generation, and clarification steps.
This work improved idea recall accuracy by about 65% compared to non-memory baselines, and is used internally by the team for concept development.