Skip to content
View gzhzk's full-sized avatar
🔆
Focusing
🔆
Focusing

Highlights

  • Pro

Block or report gzhzk

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
gzhzk/README.md

Hi, I'm Zekai Huang 👋

English· 简体中文

「Not swayed by praise, not shaken by blame; walk the path, keep myself upright.」—— DeepSeek V4

South China University of Technology | Intelligent Science and Technology | Class of 2028

On the path to AGI, Ideate, Create, Monetize.

🔭 What's Next

Goal: become an LLM algorithm engineer at a major tech company, building AI systems that actually ship.

Currently working on: Post-Training · Agent Harness · Agent Memory · RLHF

🛠 Tech Stack

Python PyTorch Linux SQL Git HTML

🔧 Tools

Claude Code Codex Hermes LLaMA Factory Weights & Biases Docker Hugging Face Jupyter

🚀 My Philosophy

Impatience comes up sometimes. There is so much to learn, and progress often falls short of expectations; watching peers publish papers, ship projects, and land offers can be unsettling. But impatience is one thing, and the path still has to be walked one step at a time.

The reality is pretty ordinary: an undergraduate in the Class of 2028, less than a year into the LLM algorithm field, still climbing in both coding ability and algorithmic thinking. On post-training, agent harnesses, agent memory, and related directions, the current state is mostly "knowing a little and filling in the gaps" — nowhere near proficient.

That said, there is one habit worth keeping: putting seemingly unrelated modules together and thinking about them as a system. Post-training, agent environments, memory, evaluation, tool use — each looks like a different problem on its own; once placed inside one system, odd little "what-ifs" sometimes surface. They are not necessarily right, and not necessarily useful right away, but I tend to write them down first, then break them into small experiments and verify them with code, data, and metrics.

I read good code when I find it, and I climb when I see higher ground. No excuses, no shortcuts. A small step every day, and the work adds up. Give it a try — something will come of it.

I still want to become an LLM algorithm engineer at a major tech company, and to slowly grow into someone who can build valuable AI systems. AGI has not arrived yet, and ASI is further still — but the questions themselves are interesting enough to be worth the time and effort.

📫 Find Me

Pinned Loading

  1. gzhzk.github.io gzhzk.github.io Public

    华南理工大学智能科学与技术专业 | AI 时代,更需要 new ideas

    HTML 1

  2. nanodeer nanodeer Public

    A reference implementation for LLM Agent runtime engineering — native ReAct loop, sandbox isolation, flat-file memory, SSE streaming API

    Python 2

  3. alignsql alignsql Public

    Qwen3-8B NL2SQL post-training from SFT to RL

    Python 5

  4. alignreason alignreason Public

    Post-training experiments: using reasoning traces to push 4B model reasoning ceiling on LiveBench.

    1

  5. qwen-tiny-reasoning qwen-tiny-reasoning Public

    Python 1

  6. hermes-agent hermes-agent Public

    Forked from NousResearch/hermes-agent

    The agent that grows with you

    Python 1