周子航

周子航 Zihang Zhou

清华大学博士生 | NMR AI | 模型微调 | 科研工具开发 Ph.D. Student at Tsinghua | NMR × AI | LLM Fine-Tuning

你好，我是周子航 👋Hello, I am Zihang Zhou 👋

关于我

我目前在清华大学生命科学学院薛毅教授实验室（Rainbow Lab）攻读博士学位，本科毕业于南方科技大学生命科学专业。研究方向聚焦 NMR 与人工智能。现阶段主要工作是微调 NMR pulse program 生成模型，并建立可靠的 benchmark，评估 AI 生成脉冲程序的质量。

我也持续开发 AI 辅助科研工具，常用技术包括 Python、Bash、Docker 与 Linux。做工程时，我更看重代码质量、可复现性和稳健交付。

研究方向

  • NMR pulse program 生成：构建和微调面向 NMR 脉冲程序的生成模型
  • 模型评估：围绕 9 个测试用例建立 benchmark，并使用 NMRSIM 验证生成质量
  • 科研数据工程：制作高质量 NMR pulse program 微调数据集
  • AI 科研工具：开发帮助研究者生成、解析、比较和验证脉冲程序的工具

荣誉与竞赛

  • 优秀奖 · 第一届全球大学生生命科学挑战赛（华北赛区 · 研究生组 · 赛道二：AI 基因编辑与核酸工具设计），2026 年 9 月
    • 项目：AI驱动的可解释RNA能量模型及其在核酸工具设计中的应用
    • 团队成员：康健宁（队长）、周子航、袁雅萱、时钰
    • 核心工作：融合纯 Python Turner 2004 热力学引擎与基序级神经网络修正，解决近简并构象辨识与变温外推难题，应用于 RNA 温度计等温度响应型核酸工具设计。
    • 链接：代码仓库 · 赛事官网

项目

AI NMR Pulse Sequence 生成

nmr.201014.xyz

用 AI 生成并评估 NMR pulse sequence，目标是提升脉冲程序研发效率和结果可靠性。

NMRagent —— 面向 NMR 的 AI Agent

nmragent.201014.xyz · GitHub 仓库

帮助 NMR 研究者生成、解析、比较和验证脉冲程序的 Agent 智能助手与服务平台，结合 70+ 份领域手册构建的结构化知识库与 RAG 检索。

GraphPulse —— NMR 脉冲序列时序图可视化

graphpulse.201014.xyz · GitHub 仓库

TopSpin 风格的现代交互式 NMR 脉冲程序可视化查看器，基于 AST 语义解析将 Bruker 脉冲程序渲染为多通道并发泳道，并支持代码语句与时序事件联动高亮。

AI 驱动的可解释 RNA 能量模型（竞赛获奖项目）

GitHub 仓库 · 赛事官网

第一届全球大学生生命科学挑战赛优秀奖项目（团队：康健宁、周子航、袁雅萱、时钰）。构建可解释 RNA 能量模型，通过最近邻基序修正与物理双链表示优化，实现高精度 RNA 自由能计算与构象切换理性设计。

开源项目

我主要维护以下 NMR AI、RNA 与 Python 开源项目：

  • GraphPulse（在线服务）：TopSpin 风格的 NMR 脉冲序列时序图交互式可视化工具
  • NMRagent（在线服务）：面向 NMR 研究者的 AI Agent 智能助手
  • evalpulse：兼容 DeepEval 的 NMR 脉冲程序多级评测基准与语义打分套件
  • pp2ast：Bruker NMR 脉冲程序语法解析器与抽象语法树（AST）构建工具，支持结构分析与 LSP 语言服务器
  • nmr-posttrain：NMR 脉冲程序大模型后训练套件（SFT + RL/GRPO），统一管理课程微调与物理验证
  • topspincli：Bruker TopSpin 命令行操控与批处理/NMRSIM 自动化工具链
  • topspinRE：Bruker 脉冲程序逆向工程与 AST 结构化解析工具
  • competition：AI 驱动的可解释 RNA 能量模型与核酸工具设计
  • spinachpython：NMR 语义分析与 Python 工具集
  • pp2label：脉冲程序标注与微调数据工程
  • prodopy-wt：Python 科学工程工具

其他作品

技能

  • 编程与工程：Python、Bash、Docker、Linux、Git
  • AI 与模型：大语言模型微调、数据集构建、Agent 工具开发、模型评估
  • 科研方向：NMR pulse program、脉冲序列生成、NMRSIM 验证

兴趣

工作之外，我喜欢小说创作、钢琴、游泳。影音上是重度科幻/机甲爱好者：《星球大战》系列、《攻壳机动队》《新世纪福音战士》《爱，死亡和机器人》《黑客帝国》，漫威宇宙几乎全看，动画追番清单也很长（《JOJO 的奇妙冒险》《咒术回战》《间谍过家家》《Re:0》等）。游戏偏爱高操作战斗与电影化叙事，《艾尔登法环》《赛博朋克 2077》都通关得很投入。歌单常驻 Starset 和泽野弘之。

联系我

全网统一网名：星灵技术

About Me

I am currently a Ph.D. student in the School of Life Sciences at Tsinghua University, advised by Prof. Yi Xue in the Yi Xue Laboratory (Rainbow Lab). I received my B.S. in Biological Sciences from Southern University of Science and Technology (SUSTech). My research is focused on the intersection of Nuclear Magnetic Resonance (NMR) and Artificial Intelligence. Currently, my primary work revolves around fine-tuning domain-specific LLMs for NMR pulse program generation and establishing rigorous benchmarks to quantitatively evaluate AI-generated pulse sequences.

I also continuously engineer AI-assisted scientific tools with a technical stack centered around Python, Bash, Docker, and Linux, prioritizing code quality, reproducibility, and robust delivery.

Research Interests

  • NMR Pulse Program Generation: Developing and fine-tuning domain-specific large language models for scientific pulse sequence generation.
  • Model Evaluation & Benchmarking: Building multi-tier benchmarks (observable coarse matching, strict scoring, simulation coverage) and validating generative quality using NMRSIM simulation.
  • Scientific Data Engineering: Curating and auditing high-quality NMR pulse program datasets for domain alignment.
  • AI Scientific Tooling: Building interactive tools and AI agents to generate, parse, compare, and verify pulse sequences.

Honors & Competitions

  • Excellence Award (优秀奖) · 1st Global College Students Life Science Challenge, September 2026
    • Division: North China Division · Graduate Track · Track 2: AI Gene Editing & Nucleic Acid Tool Design
    • Project: AI-Driven Interpretable RNA Energy Model and Its Application in Nucleic Acid Tool Design
    • Team Members: Jianning Kang (Lead), Zihang Zhou, Yaxuan Yuan, Yu Shi
    • Core Highlights: Built a pure Python Turner 2004 nearest-neighbor engine coupled with a motif-level neural residual network (MotifCorrectionNet), substantially improving near-degenerate decoy discrimination (60.34% → 80.28%) and enabling rational design of temperature-responsive RNA switches and thermometers.
    • Links: Code Repository · Official Site

AI NMR Pulse Sequence Generation

nmr.201014.xyz

An end-to-end platform for generating and benchmarking NMR pulse sequences with AI, aiming to boost research efficiency and synthesis reliability.

NMRagent —— AI Agent for NMR Spectroscopy

nmragent.201014.xyz · GitHub Repository

An intelligent agent assistant and web platform for NMR researchers, supporting pulse sequence generation, parsing, comparison, and verification backed by structured RAG retrieval over 70+ technical manuals.

GraphPulse —— Interactive Pulse Sequence Diagram Visualizer

graphpulse.201014.xyz · GitHub Repository

A TopSpin-inspired modern interactive viewer for NMR pulse programs. Powered by AST semantic parsing (PpAstParser), it renders concurrent multi-channel lanes (f1, f2, grad, delay, acq, meta) with bidirectional source-to-event highlighting.

AI-Driven Interpretable RNA Energy Model (Award-Winning Project)

GitHub Repository · Official Site

Award-winning project from the 1st Global College Students Life Science Challenge (Team: Jianning Kang, Zihang Zhou, Yaxuan Yuan, Yu Shi). Constructs an interpretable RNA energy framework using nearest-neighbor motif residual learning and physical duplex modeling for high-precision free energy computation and dynamic conformation switching.

Open Source Projects

I actively maintain the following NMR AI, RNA, and scientific software repositories:

  • GraphPulse (Live Service): Interactive TopSpin-style NMR pulse sequence timing visualizer
  • NMRagent (Live Service): Domain-specific AI Agent assistant for NMR spectroscopy
  • evalpulse: DeepEval-compatible evaluation benchmark & semantic testing suite for NMR pulse sequences
  • pp2ast: Bruker NMR pulse program parser into structured ASTs with tree-sitter support, visualization & LSP IDE integration
  • nmr-posttrain: Pulse program LLM post-training suite (SFT + RL/GRPO) for autonomous NMR pulse sequence design
  • topspincli: Command-line harness for Bruker TopSpin automation, batch processing & NMRSIM simulations
  • topspinRE: Reverse engineering & AST parsing toolkit for Bruker TopSpin pulse programs
  • competition: Interpretable RNA energy model & nucleic acid tool design
  • spinachpython: Python tools for NMR simulation, semantic parsing & analysis
  • pp2label: Pulse program annotation & dataset curation
  • prodopy-wt: Scientific Python engineering utilities

Other Creative Works

Technical Skills

  • Programming & Infrastructure: Python, Bash, Docker, Linux, Git
  • AI & Deep Learning: LLM Fine-Tuning (SFT / DPO), Dataset Construction, AI Agent Engineering, Benchmark Evaluation
  • NMR Domain: Bruker TopSpin, Pulse Program Architecture, NMRSIM Simulation & Validation

Personal Interests

Outside research and engineering, I enjoy writing science fiction, playing the piano, and swimming. I am a dedicated sci-fi and mecha enthusiast (Star Wars, Ghost in the Shell, Evangelion, Love, Death & Robots, The Matrix, Marvel Universe, Jujutsu Kaisen, Spy × Family, Re:Zero). In gaming, I gravitate toward combat mechanics and cinematic storytelling (Elden Ring, Cyberpunk 2077). My playlists feature Starset and Hiroyuki Sawano.

Contact Me

Online alias across all platforms: 星灵技术 (Xingling Tech)