Lunar New Year arrived rather late this year, indirectly delaying my 2025 review.

Sitting at my desk with a mixture of anxiety, regret, and confusion, I worked through last year’s diary and tried to recover its memories. I discovered far too many mistaken decisions.

Overall, 2025 fell somewhat short of expectations:

Although I graduated successfully with a Hong Kong master’s degree, failing to extract its full value was an absolute strategic mistake. On the positive side, it did broaden my perspective, including a deeper understanding of LLM fundamentals and blockchain, though I failed to convert that knowledge into concrete outcomes.
At work, I thought more deeply about the global-sales business, the connection between myself and the business, technology-driven progress, and AI enablement. I can feel that my understanding of work has advanced again compared with three years ago.
Regarding the future, I now understand my abilities more clearly. I used to want to chase every trend: self-media, videos, selling courses, and everything else. After flailing around for a long time, I realized that my skill tree simply does not point in that direction, so I have finally let it go. Concentrating on writing suits me better.

As usual, I will begin with a detailed review of 2025.

Progress on My 2025 Goals

At work, organize the full business chain and knowledge of cross-border transactions and fund fulfillment, and write one or two documents. Essentially complete. I spent substantial time documenting transaction flows, fund accounting, and other business logic. This should become a meaningful asset for my future.
In development, complete my approach to hot-spot systems under high concurrency and write two to four documents. 20%. At the end of 2025, I did read many relevant articles, including material on consistency and performance in concurrent scenarios, but I never turned the learning into documents.
Technically, master the foundations from ML through DL, NLP, and LLMs; build a mature product with Spring AI or LangChain; and write one or two documents. 40%. I produced one document on LLM fundamentals but did not create a mature product.
In education, earn my Hong Kong master’s degree. 100%. I graduated on October 15. A year of master’s study, IELTS preparation, and applications now feels like a dream.
Professionally, reach the next level, measured through performance. With no other major distraction, go all in at work. Basically complete. My S1 result was good, and my annual rating should be at least 3.5+.
In personal finance, average a 3% annual return, specifically by converting part of my funds into US dollars for fixed-term deposits. I was cautious about the unknown. The renminbi depreciated in the second half, while most of my principal remained in the bank, leaving a miserable return below 2%.
In mindset, focus on one or two important things and stop scattering my attention. Put concrete plans into action immediately. Looking back, this goal was too vague to have much value.
In routine, sleep at 12:30 a.m. and wake at 8:30 a.m. A complete 0%. My schedule was terrible and even affected my energy during the day.
In exercise, run 700 km during the year. 90%. I ran 650 km; the target was slightly misjudged.

Reviewing 2025

There were only two main missions in 2025: graduate successfully from UST and produce better results at work.

2025 was the Year of the Snake. On New Year’s Day, I hiked Hong Kong’s Snake Hill with Brother Ji. I no longer remember exactly where it was, but we had a good time. January always has the highest forgetting coefficient, so I recall only being surrounded by chores: buying New Year supplies, arranging a Hong Kong bank card, and worrying about my younger brother’s affairs. A business trip to Hangzhou at the end of the month at least gave me a few quiet days before the holiday.

Then I went home for New Year. The 2025 Spring Festival was hardly relaxing: as soon as I arrived, every kind of matter began revolving around me, and many things I had planned for the holiday went unfinished. At first, I was deeply frustrated by my failure to follow the plan, but gradually I became numb and silently permitted myself not to finish it—which is not a good outcome. During the holiday, I thought: I often attribute flaws in my personality to family education, but after thirty that may no longer be a good excuse. By then, I will have spent half my life away from home and must bear the cost of every outcome myself. The middle and latter parts of February were extremely difficult, chiefly because of the Batan PayPal project. Going from a beginner to launching the entire chain—from merchant onboarding and payment through settlement and accounting—nearly overwhelmed me. Yet I also gained a great deal from it.

In March, I reconsidered my direction and hoped to shift the center of my work toward AI while keeping up self-media operations. I did little concrete work, however. Looking back from 2026, I give the decision itself full marks, but my execution after March lagged badly. Other than bringing my WeChat public account close to one thousand followers, neither initiative produced substantial progress, especially AI.

March and April brought not only work but weekly travel between Hong Kong and Shenzhen for classes. The large-model and NLP courses nearly broke me. As a beginner who could not even configure a Python environment, I suddenly had to run LLM inference in Python. During early assignments, almost half my time went into learning prerequisites and configuring the environment. Worse, I ran the code on a Thunderobot laptop bought in 2017. Its fan mixed with the hot cicadas outside and repeatedly made me restless. Fortunately, I could use the company’s GPU resources for free. Thanks to Alibaba and Bailian, I obtained a 32-core GPU on weekends to finish the assignments. In the process, I learned how to perform SFT on a model and about model classification. The gains were substantial.

May was filled with deadlines for projects across all my courses. I spent almost every day on assignments and, for the first time, felt that I might not withstand the load. Staying up late every night made it impossible to focus all my energy on work. When the assignments ended in late May, I fell back into indecision: what was my career plan? Should I take advantage of studying in Hong Kong to remain there for work, or go to Singapore?

From June through September, I shifted my attention back to work. Several major events occurred:

The department’s collaboration model changed. Our group brought in many outsourced employees, and I had to coordinate with and train two or three of them while completing my own work. During this period, I worked intensely from 9:50 a.m. to 10:30 p.m. almost every day and felt physically drained. Rather than passively enduring it, I proactively asked my manager to clarify how many person-days belonged to outsourced work and loosen the allocation for my own business. This gave me room to breathe and let me train the outsourced colleagues more effectively to accept requirements.
My business scope also changed. Because another colleague planned to leave, I additionally took over fund-related requirements covering payments and settlement. It was challenging, but the new scenarios taught me more business knowledge. This was one important reason I earned a good S1 rating.
I began formally mentoring someone for the first time, four years after graduation. The intern was only a year younger than me but carried the unmistakably dazed air of a student. I believe I guided him attentively, and he successfully passed his conversion defense in September.
I used company travel funding for a business trip to South Korea, my first trip abroad. Shenzhen is a remarkable place. Since moving there in 2024, I have visited Hong Kong and South Korea and encountered different lifestyles and working cultures. The world feels full of possibilities.
I organized a team self-driving trip for the first time. The intern, outsourced colleagues, other teammates, and I drove Guangdong’s Route 318—an amazing journey.
I wrote several popular articles. How Programmers Should Configure a MacBook Pro produced more than 30,000 views on my WeChat public account for the first time, a small breakthrough. I followed it with articles on blockchain and the digital renminbi, which also performed well.

October brought graduation, finally closing my year as a Hong Kong master’s student. I even received the school’s Top Student Reward scholarship. At work, I entered the S1 review and received a decent result. I had wanted to use that momentum to keep charging ahead, but at the end of the month I learned that our business had been assigned to another manager, who brought several trusted lieutenants with him. The situation instantly felt complicated.

I spent the remaining time preparing again for the early-November professional software exam. The result shocked me: I passed comprehensive knowledge and case analysis, but failed the essay section I had considered certain. I was upset for a long time. It felt as if my effort had brought no return, and I even formed the illusion that no amount of work would let me succeed. My confidence took a real blow. Must everyone pass through such a stage in life? I remembered a sentence by Wang Xiaobo and the person I was on my twentieth birthday. Back then, I wanted to become a cloud in the sky. Now I seemed to have been slowly hammered flat.

I should have studied AI seriously and prepared paid courses on agent development. Curiosity distracted me instead: in late November and early December, I served as a teaching assistant for UST’s blockchain lecturer. He was Indian and had a strong accent, making communication difficult. The experience also led me to revisit blockchain and Ethereum, and overall I learned something. In hindsight, though, the path had drifted. I should have gone all in on agents then and at least completed the goal I set in 2024.

To summarize simply, 2025 had too many branching paths. I departed from my 2024 goals and seriously lost touch with AI agents, though I understood LLM fundamentals reasonably well. This affected a series of decisions in 2026.

Goals for 2026

At work, find a new direction and land it successfully, integrating AI coding and AI-assisted work fully into my job.
Technically, discover and resolve latent transaction-consistency problems in the system.
In AI, master the internals of one agent framework, such as Spring AI, LangChain, or Eino, and understand how coding agents work.
In writing, force myself to collaborate and publish two WeChat public-account articles each month. This is the most important goal.
In education, obtain the professional software certificate.
In personal finance, allocate 10% of my assets to US stocks and target a 3% return.
In exercise, run 700 km.
In health, undergo gastroscopy and colonoscopy and establish a 12:30 a.m.–8:30 a.m. sleep schedule. Stop sitting for long periods and reduce the fat around my lower abdomen.
In focus, keep phone use below two hours on more than twenty working days.