My name is Haoxuan Dong, and I also use the name Jason. I am currently studying in Melbourne, Australia, while using my time outside school for software development, AI experiments, and independent research.
I did not begin engaging seriously with artificial intelligence particularly early. It was only in September 2025 that I started using basic AI tools in a more systematic way. At the time, AI was mainly a more efficient way to learn and process information: it could explain unfamiliar concepts, help organize materials, and allow me to enter fields I had not previously understood.
By January 2026, I began using workstations built around genuine agents. AI was no longer only waiting for a question and returning a response. It could work toward a goal by handling files, calling tools, running code, and continuing a task over a longer period.
That change gradually made me realize that what interests me is not only what a model can generate, but why a complete technical system can work. The model itself is important, but the way it obtains information, uses tools, handles failure, and allows its final result to be checked is just as important to whether the system is reliable.
Moving to Australia also gave me more opportunities to engage with these technologies.
In the past, access to some international AI services and development resources was not always straightforward. Network restrictions, regional availability, and payment methods could consume a great deal of effort. In Melbourne, these barriers became much smaller. I could use leading models more naturally, read public documentation, explore open-source projects, and spend more time carrying an idea forward instead of leaving it at the stage of initial interest.
This change was important to me. It did not suddenly give me all the knowledge I needed, but it shortened the distance between “I want to try this” and “I can begin.”
When I make a project, I usually begin with an idea. I then consider whether it is worth the investment, how far my current conditions can support it, and whether it might eventually become a genuinely complete result. Some ideas are abandoned after this evaluation; others gradually become long-term projects through repeated revision.
AI tools make this process faster. When I encounter an unfamiliar field, I can quickly find an entry point. When I lack a piece of knowledge, I can first understand its basic structure and then continue exploring. Ideas that might once have stopped at the boundary of my knowledge can now often move a little further.
But I have also gradually found that being able to complete something with AI and genuinely understanding it are not the same thing.
I do not reject deep AI involvement in my projects. It can help with analysis, programming, and checking, and it can take on a great deal of repetitive work. For me, the question has never been whether AI should be used, but whether I still understand why the whole project works after using it.
If a piece of code runs but I cannot explain why it was designed that way; if a conclusion appears correct but I do not know what conditions it depends on; or if a system fails and my only response is to ask the model again, then even a result that looks complete can hardly become work for which I can truly take responsibility.
For that reason, I continually require myself to understand the main logic behind a project. I do not need to write every detail by hand, but the key judgments cannot be handed over entirely to a tool. AI can expand the range of what I am able to complete, but it should not replace my judgment about direction, quality, and correctness.
This requirement also shapes how I think about completeness.
I care a great deal about accuracy and professionalism. I do not want a project merely to “work well enough,” nor do I want its structure, language, or verification to feel rough simply because it is personal work. Whether the interface is complete, whether the writing is accurate, whether the result can be reproduced, and whether the limitations are stated honestly are all part of the work itself.
Of course, I still lack a great deal of experience. Sometimes I underestimate the complexity of a problem, and sometimes an attempted design proves unsuitable. But professionalism does not mean never making mistakes. A more practical standard is whether, after discovering a problem, I can re-examine my assumptions, change the method, and admit that certain approaches did not reach the expected result.
I also do not believe that complexity itself represents higher quality. A system using more models, more steps, or more technical terminology is not necessarily better. Sometimes the right task is to remove structures that do not help and allow the problem to become clear again.
At present, I am still exploring my future direction. I hope to study at the University of Melbourne and continue in software, computing-related fields, or medicine.
These directions may look different on the surface. What interests me about software is its ability to turn abstract ideas into systems that can run, be tested, and be improved. Medicine attracts me because it places high demands on knowledge, accuracy, and responsibility, while ultimately dealing directly with real people and real problems.
I have not yet decided which direction I will ultimately take, and I do not want to give an answer that is not yet true simply to make a personal introduction feel complete. Technology and medicine are not entirely separate either. In the future, I may choose one, or I may enter a field where the two intersect.
MidnightDesk is the personal website where I preserve these experiences.
It is not a product homepage, nor an extended résumé. The homepage and work pages show the things I have made, while the About page is a better place to keep another part of the story: how I first encountered technology, why I chose my current way of working, and how I want to treat my own work.
This website does not record a conclusion that is already complete. I am still learning, and I will continue to revise judgments I made in the past.
But at least for now, I have gradually formed one fairly certain belief: AI can help me enter fields that were once difficult to access, and it can help me realize ideas more quickly; but what must ultimately remain is an understanding of the technology itself, and the ability to take responsibility for the final result.