An educational RPG prototype built through AI-assisted development and evidence-based review.
DragonMagic connects math and English practice with an adventure: choose a stage, answer a question, and progress through a fantasy world. I use this personal project to turn product requirements into working screens, review AI-assisted changes, and decide what evidence is needed before accepting them.
Actual Godot runtime title screen, May 17, 2026. The opening screen introduces the math and English adventure.
- Define user flows, learning constraints, and concrete acceptance criteria.
- Coordinate AI-assisted implementation with ChatGPT, Claude, and Codex.
- Compare rendered screens and interaction behavior with the requested result.
- Request corrections when a completion claim lacks the evidence needed to support it.
- Maintain review notes, test checkpoints, and a clear distinction between working features and design proposals.
Godot 4.6.1 and GDScript support the prototype. My emphasis is product direction, workflow design, and verification alongside AI-assisted implementation.
| Evidence | What it shows |
|---|---|
| Runtime gallery and progress | Title screen, stage selection, the hub, and math practice |
| Verification record | Executable learner-profile checks and dated project evidence |
| Regression case study | Why an apparently safe cleanup broke stage navigation |
| Learner-profile source sample | Explicit rules for hints, timing, difficulty, and fallback behavior |
| Repository scope | Source provenance and what this public selection includes |
With Godot 4.6.1 on your path, run from this repository:
godot --headless --path . --script res://tests/profile_resolver_test.gdLatest local result: 31 passed, 0 failed on September 13, 2026. The checks exercise the unchanged source sample using synthetic profiles: fallback behavior, default selection, timing, hints, difficulty, and explicit settings. No original quiz data or accounts are needed. A GitHub Actions workflow runs the same command on pushes and pull requests.
Actual Godot runtime capture, May 18, 2026. A math question presents four answer choices and visible progress.
The prototype contains stage selection, quiz-based battle logic, separate math practice, and result screens. The last recorded development focus was a short W1-5 boss-battle slice and more consistent visual presentation. Later UI and motion experiments remain separate from the implemented game.
A scene cleanup removed nodes that looked unused in a text search. The scene still referenced them through names assembled at runtime, so stage navigation failed. The project record documents the failure, the reversal, and a successful follow-up check across 14 scenes. Read the case.
That is the standard I aim for: a change is ready when the relevant behavior has been checked and the result can be explained.
Personal prototype; selected public portfolio. The latest preserved runtime screenshots are from May 2026. A September 13, 2026 data check validated 1,324 quiz items with zero schema errors or warnings. This repository presents selected evidence and a source sample; it is not the full game distribution.
Created by Taeyoung Noh.