About

I am HE ZHU (朱赫) (CV / 中文), an M.Phil. student in Smart City and Big Data at Peking University, advised by Prof. Wenjia Zhang and Prof. Guanhua Chen. I previously received my B.E. in Computer Science from the Southern University of Science and Technology, where I was advised by Prof. Zipei Fan and Prof. Xuan Song. My research centers on large language models for urban intelligence, focusing on instruction data synthesis, supervised fine-tuning (SFT) mechanisms, and domain-specific deployment. I lead the PlanGPT series and am currently a research intern at Microsoft Research Asia (GENAI Group).

My research interests broadly lie in the areas of Machine Learning (ML), Natural Language Processing (NLP), and Large Language Models (LLM). More specifically, I focus on data-centric LLM research and domain-specific deployment. I am also currently exploring On-Policy Distillation (OPD) and Agent Harness. My research includes:

• (i) High-Quality Data Selection: AlignDiff, InstructDiff, Dripper, Topic Over Source
• (ii) Scalable Data Synthesis: FANNO, Tag-Instruct
• (iii) Fine-Tuning Mechanisms: ASFT, CFT
• (iv) Domain-Specific Foundation Models (Urban Intelligence): PlanGPT, PlanGPT-VL

📰 News

Education

• 
  Peking University — M.S. Smart City & Big Data, Advisor: Prof. Wenjia Zhang (Sep 2024 – Present).
• 
  Southern University of Science and Technology — B.E. Computer Science, Advisor: Prof. Xuan Song (Sep 2020 – Jul 2024, GPA 90.2/100, top 10%).

Projects

I am the first author and technical lead of PlanGPT, the first systematic study of LLMs for urban planning. The suite includes:

Learn more: plangpt.github.io

• PlanGPT — Text LLM for planning scheme generation, powering Urban-Thinking, CoPlanner, and AuditPlanner.
• PlanGPT-VL — Vision-language model for planning maps with hallucination control and hybrid training. Outperforms larger general VLMs on PlanBench-V.
• PlanGPT-R1 — Reasoning-enhanced model for long-form planning documents.
• UP-Bench & PlanBench-V — Benchmarks covering text, vision, and multimodal planning tasks.
• UrbanClaw — AI-powered urban planning assistant with multi-agent collaboration, tool-use, and vision capabilities. app.urbanclaw.net

First Author Paper

• Anchored Supervised Fine-Tuning ICLR 2026
  HE ZHU*, Junyou Su*, Peng Lai*, Wenjia Zhang, Linyi Yang, Guanhua Chen✉
• InstructDiff: Domain-Adaptive Data Selection via Differential Entropy for Efficient LLM Fine-Tuning ACL 2026 (Main)
  Junyou Su*, HE ZHU*✉, Guanhua Chen✉
• PlanGPT: Enhancing Urban Planning with Tailored Language Model ACL Industry 2025 · Oral
  HE ZHU, Guanhua Chen, Wenjia Zhang✉
• Augmenting High-Quality Instruction Data with Open-Sourced LLMs Only ACL 2025 (Findings)
  HE ZHU, Yifan Ding, Yicheng Tao, Zhiwen Ruan, Yixia Li, Wenjia Zhang, Yun Chen, Guanhua Chen✉
• Tag-Instruct: A Scalable Framework for Controlled Instruction-Tuning Data Synthesis ACL 2025 (Findings)
  HE ZHU, Zhiwen Ruan, Junyou Su, Xingwei He, Yun Chen, Wenjia Zhang, Guanhua Chen✉
• PlanGPT-VL: Vision-Language Model for Urban Planning Maps EMNLP Industry 2025
  HE ZHU*, Junyou Su*, Minxin Chen*, Yun Chen, Guanhua Chen, Wenjia Zhang✉
• Personalized Individual Trajectory Prediction via Meta-Learning SIGSPATIAL 2022 · Oral
  HE ZHU, Liyu Zhang, Zipei Fan
• Towards Fair and Comprehensive Evaluation of Routers in Collaborative LLM Systems ACL 2026 Submission
  Wanxing Wu*, HE ZHU*, Yixia Li*, Yun Chen, Guanhua Chen✉
• AlignDiff: Exploiting Model-Intrinsic Information for Better Data Selection ICML 2026 Submission
  Peng Lai*, HE ZHU*, Zhiwen Ruan, Dongdong Zhang, Yun Chen, Peng Li, Furu Wei, Yang Liu, Guanhua Chen✉

Select Co-authored Paper

• Enhancing Large Language Model Reasoning via Selective Critical Token Fine-Tuning
• Dripper: Token-Efficient Main HTML Extraction with a Lightweight LM
• Topic Over Source: The Key to Effective Data Mixing for Language Models Pre-training

Experience

• 
  Microsoft Research Asia · GENAI Group — Research Intern, Prof. Li Dong (Oct 2025 – Present).
• 
  Shanghai AI Laboratory · OpenData Lab — Researcher, foundation language models (May 2025 – Oct 2025).
• 
  SenseTime · Foundation Language Model Center — Researcher, foundation models (Jun 2024 – Sep 2024).
• 
  LocationMind, Tokyo — Project Assistant, autonomous driving algorithms (Jun 2023 – Dec 2023).

Research Appointments

• 
  SUSTech-NLP Group — Research Assistant, Prof. Guanhua Chen (Nov 2023 – Present).
• 
  Center for Spatial Information Science, UTokyo — Research Assistant (Feb 2022 – May 2022; Jul 2023 – Sep 2023).
• 
  School of Computing, NUS — Visiting Student (May 2022 – Jul 2022).

Honors & Service

• Outstanding Youth League Secretary, Peking University — led the branch to Five-Star recognition (2025)
• Outstanding Graduate & Outstanding Thesis, Dept. of CS, SUSTech (Top 5%, 2024)
• Annual Outstanding Student, SUSTech (2021, 2022, 2023)
• Reviewer: ARR (ACL Rolling Review)