Rebuilding My Personal Site Around a Clearer Thread

I do not want this site to become a pure academic homepage.

That is not the right shape for the current stage. This site is closer to a technical blog, a project page, and a curated reading shelf. I will use it to write about systems I have built, papers I have read, bugs I have debugged, and questions I have not fully figured out yet.

The more important goal is to make my own thread clearer.

Over the past period, I have worked on a lot of Agent and AI Infra productization work: Agent Runtime, MCP tools, sandboxed execution, OpenAPI Gateway, authentication and billing, rate limiting, tracing, prompt evaluation, and memory services. Those can sound like a pile of technical terms, but the real lesson is simple: once an agent moves from demo to product, it is no longer just a model call with a few tools attached.

It becomes an engineering system.

Task state needs to be managed. Tool protocols need to be stable enough for models to consume. Execution environments need isolation. Failures need traces. Bills need to reconcile. Prompt changes need regression tests. Memory cannot just mean putting chat history into a vector store.

These experiences now shape how I look at research problems. Coding-agent post-training, long-horizon task synthesis, sandbox construction, verifier design, SFT/RL data quality, and credit assignment may sound more algorithmic, but the substrate is still deeply engineering-heavy.

What kind of data trains long-horizon coding agents? What kind of environments and verifiers can support reliable RL? When a model fails in a long task, should we blame the model, the data, the tools, the environment, or the task definition itself?

These are not problems that can be solved by simply running more turns.

So the first version of this site stays simple:

• Blog is for my own writing: engineering retrospectives, research notes, and debug logs.
• Curated is for external resources I actively select and annotate.
• Projects is for systems, papers, and portfolio-level work.
• About explains who I am and why I care about these problems.
• Contact keeps the public ways to reach me.

The old Notes/Archive content was not mine, so the first public version removes it directly. At this stage, it is more important to put my public writing into Blog than to split posts, notes, and research fragments too early.

I want the writing here to be real, restrained, technically grounded, and reflective. Not “a complete guide to best practices”, and not an AI-flavored brochure. The better version is: why I was confused, why the problem appeared in real engineering, how I initially misunderstood it, how I debugged it, which designs were trade-offs, and which parts I still have not fully figured out.

In one sentence, this site is not about a student with many scattered experiences. It is about a gradually forming thread.

In the past, I worked on runtime, tools, gateways, memory, observability, and prompt evaluation in real agent products.

Now, I am bringing that systems experience into coding agents, LLM post-training, long-horizon task synthesis, and executable evaluation.

Long term, I want to build AI-native systems that can become both papers and products.