
AI Meeting
AI models are improving rapidly, but progress in model capabilities has not automatically translated into equivalent value at the application layer. A model that can answer questions is not the same as a product that consistently delivers value in real workflows.
I believe the next opportunities go beyond building stronger models. They lie in connecting AI with domain knowledge, business data, and existing workflows — enabling agents to understand tasks, use tools, execute processes, and collaborate with people when needed.
I'm a Computer Science student at Taiyuan University of Technology, focusing on AI products and applications. Through hands-on projects, I want to better understand this process: identifying real problems, designing useful systems, and evaluating the outcomes they produce.
Start with the user's problem, workflow, and practical constraints.
Connect models, domain knowledge, tools, and human oversight where needed.
Evaluate task quality, time, cost, and user feedback — not just model performance.
The real value of AI ultimately has to show up at the application layer.
NOW — 2006
AI product form · User growth · AI education
I am currently exploring AI product forms and user growth, and doing AI science communication and teaching talks. I am only a sophomore, but I have extremely high enthusiasm for AI and hope to learn and exchange with more people.

Open-source model router · Cost per solved task
On 19 November 2026, I built ModelPick for the need to choose different models by task complexity and cost. Rooted in Claude Code, OpenAI, Cursor and other ecosystems, it improves model selection for complex tasks, lowers cost, and raises cost-effectiveness and task efficiency. It is my latest project on GitHub and has been announced on X; feedback is welcome.

Shanghai hackathon · Claw Hunt · AI game co-creation
In November 2026, we came to Shanghai and joined the Shanghai hackathon at the invitation of Claw Hunt. Create World became the world’s first AI game creation platform and open-source game co-creation community; it is now on my GitHub and already has 100 core users, and is still being iterated.

Insta360 Imaging Challenge · Public-welfare track
In October 2026, I was invited to the Insta360 Imaging Challenge — my first hardware-software hackathon. Our product was a portable child early-screening imaging device; it sparked broad discussion in the public-welfare track and drew attention to early disease screening for infants and young children.

LLM application development intern
In September 2026, I joined Cofoe Technology as an LLM application development intern. Within one month I completed the whole process from requirement breakdown to deployment and trial, creating more than ¥150,000 in delivery value for the company.

Douyin AI Innovator regional · Wuhan
In July 2026, we advanced to the Douyin AI Innovator regional in Wuhan. Our product PocketMentor is also on GitHub; at the event we received broad recognition from the judges and audience.

Douyin AI Innovator Shanxi / China Robot and AI Competition / China University Intelligent Robot Creative Competition
From March to May 2026, I started doing hackathons. The first was the 2026 Douyin AI Innovator Shanxi regional, where we won first prize; the first-prize project was Garden of Suzhou. We also took part in two national A1 competitions — the China Robot and AI Competition and the China University Intelligent Robot Creative Competition — and won national-level awards in both. The crowd-simulation project is also on GitHub.

Deep learning & diffusion models
From October 2025 to February 2026, I studied deep learning principles and diffusion models systematically at the TYUT-MIHRI lab, and tried to reproduce Shine, a recent diffusion-model study based on Flux. This laid a solid theoretical foundation for my later understanding of AI models and learning of AI products.

Computer Science and Technology
In September 2025, after the Chinese Gaokao, I entered the Computer Science and Technology programme at Taiyuan University of Technology. I think my university is a very good 211 university in China, and many academicians of the national academies of sciences and engineering grew up here. I am grateful to my university for giving me a broad platform to grow.

Pingyao, Shanxi · High school
From September 2022 to June 2025, I studied at Pingyao High School, the best high school in Pingyao. It gave me a solid foundation and a rigorous way of thinking in the sciences; its motto — resolute, united, practical, innovative — has stayed with me, and it is where my knowledge and learning began to grow.

Pingyao, Shanxi
I was born in Pingyao County, Shanxi, on 7 November 2006. I love my hometown. Pingyao carries an ancient Chinese culture, and it also shaped a strong and steadfast inner self in me.

Five departments, five kinds of meetings, one noisy room. The job was not to transcribe audio; it was to turn a meeting into a record that someone could approve.
I owned the product: decomposing requirements, choosing a two-stage ASR pipeline, defining the human check, and getting the thing deployed on a factory intranet.
The pipeline
Three decisions
Turned five department scenarios into one priority map.
Two-stage ASR: realtime during the meeting, accurate after it.
Designed the approval path, exception recovery, and intranet rollout.
Seven projects across product, agents, and creative web. The interesting part was never the demo.

Routes every task to the model that solves it at the lowest cost per solved task.

Turn a story idea into a playable pixel game — then publish and remix it.
An enterprise meeting assistant that turns a noisy room into an approved record.
A slow-brain / fast-brain agent simulation for disaster response.
A generative 3D garden where AI assets and 2D/3D rendering share one space.
A real-time video expert that helps restaurant founders make better decisions.
A portable first-person device that looks for early developmental signals at home.
See more of my AI work.
Questions about AI products, agents, competitions, or what I am building next. I read everything.
If you are building an interesting AI product, I would like to hear about it.