About me


This is where you learn some basic things about me

Education

Nanjing University, Suzhou, Jiangsu
Undergraduate School of Intelligent Science and Technology (2023 – Present)

  • GPA: 4.43 / 5.00
  • Major Courses: Fundamentals of C Programming (95), Advanced Programming (92), Introduction to Artificial Intelligence (96), Data Structures and Algorithms (98), Operating Systems (89), Introduction to Machine Learning (89), Introduction to Databases (94)
  • Honors: Third-Class People’s Scholarship (2025)

Professional Skills

  • Programming Languages: C/C++, Python, Lean4, Java, SQL, HTML/CSS, familiar with Linux development environment
  • Familiar with PyTorch framework, Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) fine-tuning for Multimodal Large Language Models (MLLMs), and experienced in building agent systems
  • Languages: English (TOEFL 111; CET-6 610; CET-4 659)

Research Experience

NeSyGeo: A Neuro-Symbolic Framework for Multimodal Geometric Reasoning Data Generation

Mar. 2025 – May 2025
Co-First Author, ICML 2025 AI4Math Workshop

We proposed a neuro-symbolic framework, NeSyGeo, which enhances data quality through symbolic generation and multimodal synthesis.

  • Designed a domain-specific language for geometry to precisely describe geometric figures
  • Built a data synthesis pipeline and utilized the synthetic data for SFT and RL fine-tuning of MLLMs, enhancing their geometric reasoning capabilities

Mind the Gap to Trustworthy LLM Agents: A Systematic Evaluation on Constraint Satisfaction for Real-World Travel Planning

Sep. 2025 – Nov. 2025
Co-First Author, AAAI 2026 TrustAgent Workshop

We established a unified evaluation framework to systematically reproduce and evaluate mainstream LLM agents in travel planning tasks:

  • Reproduced various LLM-based agent systems within a unified framework, reviewed existing benchmarks, and extracted design principles
  • Analyzed the trade-offs between generality and specialty of different agent architectures, identifying challenges in real-world scenarios

High-Fidelity Controllable 3D Digital Human Modeling

Nov. 2024 – Nov. 2025
Project Leader, National College Students’ Innovation and Entrepreneurship Training Program

We built a data generation pipeline for high-fidelity controllable 3D digital humans to synthesize large-scale high-fidelity digital human data.

  • Leveraged open-source datasets to synthesize high-quality human video data with 3D structural priors for fine-tuning the Wan2.2 model
  • Built an image synthesis pipeline with photorealistic priors and efficient inference capabilities to efficiently synthesize high-quality datasets

Neuro-Symbolic Based Embodied Task Planning in Open Environments

Nov. 2024 – Nov. 2025
Core Member, National College Students’ Innovation and Entrepreneurship Training Program

We proposed a general framework for robot manipulation in open worlds, effectively bridging the gap between high-level semantic planning and low-level motion control, significantly improving the robot’s generalization ability and task success rate in unseen scenarios.

  • Proposed a semantics-action synergistic framework, introducing structured verb tuples as a bridge between semantics and actions
  • Utilized various Vision-Language Models to generate fine-grained, intent-driven part-level heatmaps for precise perception
  • Employed Diffusion Policies to generate robust and smooth trajectories

Awards

  • First Prize, “CCB Cup” Jiangsu College Students Innovation Competition (2025) — Aug. 2025
  • Silver Award, China International College Students’ Innovation Competition (2025) — Dec. 2025

Hobbies

  • 🎸 Playing the guitar
  • 🎭 Stand-up comedy (Xiangsheng)
  • 🪄 Magic tricks
  • ⚽ Playing football