2025 Q1–Q3 notes

Another month has passed since my last update. 2025 has been busy, and often I have not had the energy to write much beyond a few casual notes. Fortunately, the products I work on have been doing well lately, so the time spent iterating on them has not been wasted. With a rare quiet weekend, I decided to clear out my photo library and look back on the first three quarters of the year.

A trip to Beijing

I traveled to Beijing for work in February. I watched the flag-raising ceremony at Tiananmen Square, ate Peking duck, drank douzhi, played basketball in Dongdan, and met friends I had not seen in a long time. After the game, we drank beer and wandered along Gulou Street talking nonsense. It felt like university again, and I relaxed completely.

A diamond ring and a proposal

I returned to Madrid during the May Day holiday. As my relationship with Pingping had become increasingly steady, I began planning a proposal in February. Friends and colleagues were very supportive; some colleagues even went ring shopping with me. After learning far more than I expected about diamond rings and deliberating for a long time, I chose Chow Tai Fook, the brand I knew best. To record the proposal, I bought an Insta360 GO and told Pingping it was simply for documenting everyday life.

On May 3, 2025, after a light rain in Plaza del Corrillo in Salamanca, we were strolling through the square. At a scenic spot, I casually took out the camera for what I claimed would be a selfie, then awkwardly pulled out the ring from my pocket and got down on one knee, just like in a TV drama. It was a little clumsy, but it still moved her to tears.

I think it will become an anchor point in our memories for decades to come.

Back in Spain

Madrid seemed unchanged from my previous visits, but walking its streets felt very different from my student days. After entering the working world, I have a different understanding of life in Europe.

The itinerary was packed. I met several old friends, each now on a different path and in a different industry. I also attended a friend’s Western-style wedding and danced with everyone there—an experience I rarely get to have.

Learning to drive

It is a little embarrassing that I only started learning to drive at this age. I had played plenty of racing games—Forza Horizon, Gran Turismo, and Need for Speed—but still made mistake after mistake behind the wheel. Practice was intermittent, and I was so nervous during my first test for the parking-and-maneuvering section that I crossed the line twice while parallel parking. I finally passed a month later.

Switch 2

I bought a Switch 2 in July. I had owned the original Switch too, but it gathered dust once the novelty wore off. When the Switch 2 launched, the constant coverage from media creators eventually wore me down; after holding out for two months, I placed an order.

At first, it brought back the excitement I used to feel for games. I spent more than ten hours on Pokémon, then another ten-plus hours deep into Dave the Diver. I could not get into the rest—whether Luigi’s Mansion or Mario Kart World. I am playing P5S on and off now, but it looks as though this console may gather dust too.

A trip to Jinan

At the beginning of August, Pingping came to China for company training after a job change, so we were able to meet again. It had been nearly two years since I returned to China. For a cross-border relationship like ours, these occasional short reunions seem to have become normal.

We visited most of Jinan’s limited number of attractions. The most memorable part was the sea of people at Baotu Spring. August was still intensely hot, and walking around was genuinely exhausting.

Recently, we have also developed the odd hobby of mixing drinks with convenience-store ingredients. So Jinan’s streets gained two more bartenders.

Work

Finally, work. Over the first nine months of the year, I gradually worked on feature engineering and semantic mining. In particular, I made some progress with pretrained language models and ID-sequence features for semantic mining, with modest gains in both offline tests and online A/B tests.

My colleagues’ work has been much more impressive. One colleague achieved excellent online results with an algorithm that was a major breakthrough in both architecture and modeling; others brought business ideas I had never considered.

That comparison naturally creates some technical anxiety. My colleagues built on an existing framework, while I was building one from scratch. I spent a great deal of time debugging and worked inefficiently. I have started learning from them, because business teams are ultimately driven by efficiency and return.

The next stage of my work will, as expected, remain focused on recommender algorithms—especially multi-task learning. I hope to deliver something meaningful in the next three months.

September 6, 2025, in Suzhou