GamED.AI: Framework for Automated Educational Game Generation
GamED.AI is an advanced multi-agent framework that converts questions from instructors into engaging, educational games grounded in pedagogy, verified through formal mechanic contracts. Utilizing phase-based LangGraph sub-graphs, deterministic Quality Gates, and organized Pydantic schemas, it features two template families with 15 interaction mechanics that cover spatial reasoning, procedural execution, and higher-order objectives from Bloom's Taxonomy. The system was tested on 200 questions across five subject areas, achieving a 90% validation pass rate, 98.3% schema compliance, and a 73% reduction in tokens compared to ReAct agents (from ~73,500 to ~19,900 tokens per game) at a cost of $0.46 each. Results indicate that the phase-bounded architectural structure is more closely related to alignment quality than the prompting strategy. Attendees can interact with the demonstration.
Key facts
- GamED.AI is a hierarchical multi-agent framework for automated educational game generation.
- It transforms instructor-provided questions into playable, pedagogically grounded games.
- Built on phase-based LangGraph sub-graphs, deterministic Quality Gates, and Pydantic schemas.
- Supports two template families with 15 interaction mechanics.
- Evaluated on 200 questions across five subject domains.
- Achieves 90% validation pass rate, 98.3% schema compliance.
- 73% token reduction over ReAct agents (~73,500 to ~19,900 tokens/game).
- Cost per game is $0.46.
Entities
—