ABOUT
Hello, I write as 吃栗子不. I work in algorithm engineering, mainly on recommender systems. Most of my day-to-day work involves ranking, multi-task learning, features, and evaluation. Outside work, I also experiment with LLMs, RAG, and agents.
I write things down because implementation details are easy to forget: why a metric moved, why the online result did not follow, or where an investigation finally got stuck. This blog keeps those notes, along with paper readings and code.
Currently looking at
- Turning offline gains in recommendation models into reliable online impact.
- Connecting LLMs and agents to real products rather than stopping at a demo.
- Keeping evaluation, observability, and engineering constraints level with model capability.
I only write about work that can be discussed publicly. If you are working on similar problems, feel free to reach me on GitHub or by Email.