Skip to content
View dongyuan21's full-sized avatar
🎯
Focusing
🎯
Focusing
  • beijing

Block or report dongyuan21

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
dongyuan21/README.md

English · 简体中文

Hi 👋, I'm Dongyuan

User Growth Engineer | AI Agent Builder | Olympic Weightlifter

Melo Studio live Melo source nanoAuralRuntime Data Agent


🔭 What I'm Working On

  • Casual Puzzle Games: Building Melo (source) — a match-3 simulator for importing assets, storyboarding boards, 3D rendering, and exporting video
  • User Growth: Built overseas incentive platforms from 0 to 1 — task systems, viral loops, and SEO traffic — iterated with ROI / cost / financial-loss guardrails
  • Paid UA: Running IAA user acquisition for casual games, iterating creatives against spend and ROI
  • AI Agents: A runtime-controlled, read-only Data Agent for UA — routing questions, governed SQL, and evidence-backed reports

📄 Technical Report

A Reusable Production System for Casual Puzzle Game Ad Creatives
Design, Implementation, and Production Practice
Dongyuan Li, Yuwei Liu · 2026

English PDF · 中文版 PDF

Illustrated reading editions re-typeset from the final manuscripts dated 20 September 2026. The figures are explanatory redraws; original case screenshots are not reproduced. Edition notes are included in both PDFs. These are author technical reports, not arXiv-published PDFs.

🎬 Ad Creative Pipeline

A UA creative is hook + gameplay.

The hook is batch-remade through a C-end production workflow (not in this GitHub). The gameplay body is built on melo-game-creative-studio — a simulator skeleton that imports assets, storyboards the board, renders in 3D, and exports video. Melo also ships as a CLI so agents can drive it.

When the 3D render isn't enough, Melo dumps JSON. An agent turns that JSON into After Effects / Blender projects. ae-blender-skill then parses those source files, atomizes animation elements into reusable assets, and feeds them back into Melo.

That loop exists to kill two kinds of toil: hand-tweening every shot, and manually laying out every board.

SFX for elimination games is generated by nanoAuralRuntime.

Match-3 ad creative pipeline: hook + gameplay, Melo, ae-blender-skill, nanoAuralRuntime

Open the Excalidraw source

💪 Strength Beyond Code

  • Olympic Weightlifting: Snatch 70kg | Clean & Jerk 90kg | Power Clean 100kg
  • Basketball: Team player on and off the court 🏀
  • Philosophy: "In the age of AI, aesthetic judgment and values matter most."

🌱 Currently Exploring

  • Beyond Match-3: Extending Melo's board grammar to more casual genres without rewriting the render / export loop
  • Creative Closed Loop: Feeding spend / ROI signals back into board generation and shot selection
  • Agent Batch Export: Driving Melo CLI end-to-end so one brief yields many storyboarded cuts

Pinned Loading

  1. melo-game-creative-studio melo-game-creative-studio Public

    TypeScript 1

  2. mjbcbb-iaa-data-agent mjbcbb-iaa-data-agent Public

    做一个iaa投放辅助决策的data agent

    HTML 1

  3. nanoAuralRuntime nanoAuralRuntime Public

    Python 1 1