GamED.AI
A Hierarchical Multi-Agent Framework for Automated Educational Game Generation

Live Demo | Architecture | Citation | License

---

GamED.AI is a hierarchical multi-agent framework that automatically transforms educational questions into interactive web-based games. Given any question — from "Label the parts of a plant cell" to "Trace through bubble sort" — the pipeline orchestrates specialized AI agents across 6 phases to produce fully playable games covering 15 mechanic types, 5 subject domains (Biology, Computer Science, History, Linguistics, Mathematics), and 3 education levels (K-12, Undergraduate, Graduate). The system demo includes 50 pre-generated games playable without any backend.

Citation

@inproceedings{agarwal-etal-2026-gamedai,
    title = "{GamED.AI}: A Hierarchical Multi-Agent Framework for Automated Educational Game Generation",
    author = "Agarwal, Shiven and Shah, Yash and Shekhar, Ashish Raj and Bordoloi, Priyanuj and De, Sandipan and Gupta, Vivek",
    booktitle = "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics: System Demonstrations",
    month = aug,
    year = "2026",
    publisher = "Association for Computational Linguistics",
}

Live Demo

Browse 50 pre-generated games at: https://ShivenA99.github.io/GamED-AI/

Repository Structure

GamED-AI/
├── backend/                    # Python/FastAPI backend
│   ├── app/
│   │   ├── agents/             # LangGraph agents (59+)
│   │   ├── config/             # Model registry, presets
│   │   ├── routes/             # FastAPI endpoints
│   │   └── services/           # LLM, image, storage services
│   ├── prompts/                # Agent prompt templates
│   ├── scripts/                # CLI generation scripts
│   └── requirements.txt
├── frontend/                   # Next.js frontend
│   ├── src/
│   │   ├── app/acl-demo/       # Static demo pages
│   │   ├── components/         # React components + game templates
│   │   └── data/acl-demo/      # Pre-generated game data (50 games)
│   └── public/acl-demo/        # Static assets for demo games
└── docs/
    └── ARCHITECTURE.md         # Full pipeline diagrams and agent table

---

0. Prerequisites

• Python 3.10+
• Node.js 18+
• API keys: GOOGLE_API_KEY (Gemini), OPENAI_API_KEY

---

1. Quick Start — Static Demo (No Backend Required)

Browse all 50 pre-generated games locally:

git clone https://github.com/ShivenA99/GamED-AI.git
cd GamED-AI/frontend
npm install
npm run dev
# Open http://localhost:3000 — redirects to /acl-demo

The demo runs entirely from static JSON files and local image assets — no API keys or backend needed.

---

2. Full Setup — Backend

cd backend
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Create backend/.env with your API keys:
# GOOGLE_API_KEY=your_gemini_api_key_here
# OPENAI_API_KEY=your_openai_api_key_here

PYTHONPATH=. uvicorn app.main:app --reload --port 8000
# Health check: curl http://localhost:8000/health

---

3. Full Setup — Frontend (with Backend)

cd frontend
npm install
npm run dev
# Open http://localhost:3000

---

4. CLI Pipeline Usage

Generate a new game directly from the command line:

cd backend
PYTHONPATH=. python scripts/generate_acl_demo.py \
  --query "Label the parts of a plant cell" \
  --model gemini

Output is saved to backend/assets/demo/ with the generated blueprint JSON and diagram images.

---

5. Game Library

The demo includes 50 pre-generated games spanning:

Domain	K-12	Undergraduate	Graduate
Biology	Plant cell, Mitosis, Food chains	Cardiovascular, DNA enzymes, Inheritance, Organelles	Cellular respiration, Clinical diagnosis, Evolution
Computer Science	Bubble sort, Computer parts, Network data	BFS, Binary search, Algorithm complexity, UML	DP complexity, Scheduling, TCP/UDP
History	American Revolution, Historical figures, Roman Empire	Ancient civilizations, Industrial Revolution, Revolutions, WWI causes	Cold War, Historiography, Versailles
Linguistics	Sentence parts, Sentence building, Word types	Language families, Morphemes, Phonology, Semantic roles	Language acquisition, Phonological processes, Syntax trees
Mathematics	Geometry, Linear equations, Number types	Calculus, Function types, Integration, Matrix ops	Numerical methods, Proof strategy, Vector spaces

Mechanics covered: drag-and-drop, click-to-identify, trace-path, sequencing, sorting, memory-match, branching-scenario, compare-contrast, description-matching, state-tracer, bug-hunter, algorithm-builder, complexity-analyzer, constraint-puzzle, and hierarchical.

---

6. Architecture Overview

The GamED.AI pipeline is a hierarchical DAG in LangGraph with six phases, each an independent sub-graph with typed I/O and a Quality Gate at its boundary:

1. Phase 0 — Context Gathering: Parallel input analysis and domain knowledge retrieval grounded in curated sources.
2. Phase 1 — Concept Design: ReAct agent resolves input against a Bloom's-to-mechanic constraint table to produce a game concept with learning objective, template family, and mechanic contract.
3. Phase 2 — Game Plan (deterministic, no LLM): Assigns scene IDs, computes score contracts, determines asset needs, builds transition graph.
4. Phase 3 — Scene Content (parallel): N parallel Send() calls generate game-type-specific content per scene. FOL-based Bloom's alignment predicates at QG3.
5. Phase 4 — Assets (parallel): M parallel workers perform image search, quality filtering, and fallback generation.
6. Phase 5 — Assembly (deterministic, no LLM): Combines game plan + content + assets into a verified JSON blueprint.

Four Quality Gates (QG1–QG4) execute without LLM inference, providing constant cost and formal verifiability. The architecture achieves 90% validation pass rate with 73% token reduction over ReAct baselines ($0.48/game, ~19,900 tokens).

See docs/ARCHITECTURE.md for full pipeline diagrams, phase details, and game template reference.

---

License

MIT License — see LICENSE for details.

Copyright 2026 Shiven Agarwal, Yash Shah, Ashish Raj Shekhar, Priyanuj Bordoloi, Sandipan De, Vivek Gupta