Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Former Security Guard Reaches AI Finals, High Schooler Wins 250K: TRAE AI Creativity Conference Wraps Up

TRAE's AI Creativity Conference wrapped up in Shanghai after 37,000 registrations and 14,000 submissions, with the top three projects taking home up to 300,000 yuan in prize money. A 24-year-old former security guard reached the Top 20 by building a motion-sensing health game with TRAE, while a high school sophomore's AI badminton coach hit No. 22 on the App Store hot chart within 10 days of launch.

Published
前保安杀进AI决赛、高中生拿走25万:TRAE AI创造力大会总决赛落幕
Image source: qbitai.com

The finals of the TRAE AI Creativity Conference were held in Shanghai, drawing 37,000 registrants whose 14,000 submissions were narrowed to a Top 20 before champions were crowned, with the top prize worth 300,000 yuan. Singer Hu Yanbin and Tim, founder of the video channel 影视飓风, presented awards on stage, Unitree founder Wang Xingxing joined online, and the venue was packed.

The most talked-about moment on the finals stage belonged to Peng Mingyu, a 24-year-old former security guard. He had taught himself Douyin special effects well enough to become a senior effects artist, then used TRAE to build "Neck Workout" (动动脖子) in two months — a browser-based two-player motion-sensing health game that turns your nose into a ping-pong paddle or fishing rod to loosen a stiff neck. "I'm basically illiterate when it comes to English," he said on stage, "but even with zero foundation and working alone, I built it."

In the final results, the champion project "Word Element Creation" (词元开物) took 300,000 yuan, runner-up "MotionFrame" (动作帧) won 250,000 yuan (a 200,000 yuan runner-up prize plus a 50,000 yuan social charity award), and third place "Open Source Space: Satellite Guardian" earned 100,000 yuan. Hu Yanbin and Tim personally presented Peng's award and tried the game on stage.

The champion team — full-time founders Yang Zhenyu, Zheng Zhongpeng and Lu Chenglong — targeted a clear gap in AI coding: software can be made by talking, but what about hardware? They built an AI hardware IDE paired with a desktop PCB milling machine: type "make me a music keyboard," and the AI confirms requirements, generates a PRD, selects components, auto-routes the board, and mills the finished circuit. In the semifinal round they pushed the product to roughly 500 real users, obsessing over making AI-generated output actually manufacturable, with one sprint of nearly five 14-hour days and 361 Git commits.

The runner-up was Kyle, a high school sophomore from Zhongshan, Guangdong, and his teammate Xiao Zhou. Because professional coaching was too expensive — 360 yuan per hour plus a 100 yuan court fee — Kyle started building MotionFrame on the evening of July 11: an AI badminton coach that lets you watch yourself play. The semifinal version wired together iPhone, Apple Watch and AirPods — the phone analyzes elbow angle, contact point and swing speed frame by frame, the watch captures motion and heart rate, and the earbuds talk you through corrections between rallies. The app cleared Apple review within a month, and Kyle said at the finals that it had hit No. 22 on the hot chart after 10 days on the store.

Third place went to a father-son duo: 40-year-old Tan Tan and his middle-school son Bobo, who built "Open Source Space: Satellite Guardian," a satellite situational-awareness and mission-simulation suite. New to the space industry, Tan found professional software prohibitively hard to use, so he fed TRAE a natural-language brief and let it generate a full development task document; the first complete run took just over four hours and grew to 1.63GB of code. The semifinal version added domestic AOE orbital data and a virtual "TREA-01" satellite mission center that schedules remote-sensing tasks, warns of debris approaches, generates avoidance plans, and even offers a first-person cockpit view.

Notably, contestants also included a geology student, a freshly laid-off UI designer, and the self-described "illiterate" former security guard — a large share of people who had never been traditional programmers, yet delivered more than demos: one built a geology-modeling tool that runs on real hardware, one shipped an app to the App Store, and one was already testing with real users.

That reflects how TRAE itself has evolved: when it launched in early 2025 it was closer to an AI IDE, and six months later SOLO arrived, letting users state a goal while the AI decomposes tasks, reads projects and calls tools. With Skills, Rules, MCP, custom agents, Plan and Spec filled in, development increasingly means "explain the idea and constraints, and the AI plans, edits files and runs commands." The coding barrier is receding, pushing understanding, judgment and trade-offs to the front.

The bigger takeaway from the contest is simple. Technology still decides whether a product can run, but creativity increasingly decides why it deserves to exist. The open questions are whether these winning projects reach the market and whether "everyone can create" keeps minting new developers.

Why it matters

AI coding capability is rapidly moving into the hands of non-programmers; a contest with 37,000 participants could herald a new wave of AI-built applications.

TRAEAI CodingAI Contest
Back to realtime news

Nearby Updates

All