Gaoshan Zeng / 曾高山 AI for Science builder | First-principles simulation | Materials informatics | Open-source developer I am an undergraduate student in Electronic Science and Technology at Sichuan Agricultural University with a 4.22/5.0 GPA. My work sits at the intersection of AI4S, materials science, first-principles simulation, finite-element / electromagnetic simulation, chemical experiments, and AI agents. I am especially interested in building systems that close the loop between domain knowledge, simulation, experiment, and autonomous discovery. Research interests: AI4S, AI Scientist agents, ML potentials, materials informatics, DFT/MD, high-entropy materials, microwave absorption, electrochemical sensing, and high-throughput screening. Public writing: WeChat public account 飞鸟想要飞, with 1M+ total views.

Solo, fully loaded ride from Chengdu to Lhasa; photo at Zongla Mountain, 4,150 m

Million-scale high-fidelity electrolyte MLIP dataset and structure-property analysis
Peking University Shenzhen Graduate School, visiting student / research assistant, 2026.04 - present

• Proposed an adaptive, full-stack data-construction workflow for electrolyte MLIP datasets: generative-prior composition design, pulsed-dynamics sampling, and multi-fidelity quantum labeling.
• Combined classifier-guided diffusion with pulsed nanoreactor active learning under extreme temperature / volume perturbations to capture non-equilibrium transition states and high-uncertainty OOD configurations.
• Developed physics-guided reduction through adaptive fragmentation and ALMO-EDA analysis, then paired million-scale lower-cost force / geometry labels with sparse high-accuracy energy references to reduce computational cost by about 90%.

DFR-related high-throughput screening research
Department of Artificial Intelligence, Xiamen University, 2026.02

• Worked on DFR-related research for accelerating high-throughput screening workflows.
• Focused on connecting AI models with candidate generation, simulation evaluation, and fast property filtering.
• Interested in turning high-throughput screening from a manual pipeline into a more automated and reproducible research loop.

Electrodeposited Cu foils: microstructure and mechanical properties
University of Science and Technology of China / Institute of Metal Research, Chinese Academy of Sciences

• Worked around the system studied in The effect of 2-mercaptobenzimidazole concentration on the microstructure and mechanical properties of electrodeposited Cu foils.
• Studied how 2-mercaptobenzimidazole (MBI) affects electrodeposited Cu foils through grain refinement, surface morphology, crystallographic texture, nanotwinned grains, and thermal stability.
• Followed the electrochemical mechanism behind additive-assisted Cu electrodeposition, including MBI adsorption and complexation with cupric ions.

Machine learning for high-entropy materials
KAUST, visiting student / research assistant, 2025.09 - 2026.03

• Built ML workflows for high-entropy alloys and oxides using ALIGNN, composition-based feature vectors, and JARVIS-DFT data.
• Focused on formation enthalpy, total energy, magnetic moments, and generalization under sparse high-fidelity data.
• Worked on accelerating candidate screening in large composition spaces.

Electronic-structure atlas of MoS2 nanotubes with DeepH and first-principles workflows
Westlake University, research assistant, 2025.07 - 2025.09

• Integrated VASP, OpenMX, ABACUS, Quantum ESPRESSO, CHGNet, and DeepH into automated electronic-structure workflows.
• Studied chirality-dependent bandgaps, effective masses, and direct/indirect bandgap transitions.
• Compared DeepH predictions with DFT-level calculations to evaluate accuracy and computational efficiency.

AI Scientist agents for materials discovery
DeepModeling / AI4S exploration

• Working on agentic workflows for literature understanding, hypothesis generation, simulation planning, and materials-design feedback loops.
• Interested in making scientific agents more reproducible, tool-aware, and grounded in physical constraints.

LLM-guided reverse design of high-entropy microwave absorbers
Manuscript in submission

• Developing an autonomous LLM-agent-guided framework for multi-scale reverse design of high-entropy microwave absorption materials.
• Combining domain reasoning, conditional diffusion / reinforcement-learning style generation, ALIGNN-based stability evaluation, and CST-based electromagnetic simulation.
• Goal: accelerate the discovery of high-performance, ultra-broadband microwave absorbers with physics-aware constraints.

Machine-learning-integrated electrochemical sensing for farmland heavy metals
Provincial Undergraduate Innovation and Entrepreneurship Training Program, project lead, 2024.12 - 2025.09

• Led the design of an electrochemical sensing and machine-learning integration platform for Pb²⁺ and Cd²⁺ detection.
• Built a PAC-SVR-LSTM pipeline with PCA and wavelet denoising for classification, concentration regression, and dynamic interference compensation.
• Achieved a 0.8 nM detection limit, 98.2% identification accuracy, and an R² above 0.995 for concentration quantification; the project resulted in two software copyrights.

Composite nanofiber and machine-learning heavy-metal sensing platform
National Undergraduate Innovation and Entrepreneurship Training Program, core member, 2024.06 - 2025.09

• Prepared composite nanofiber electrodes through electrospinning and optimized electrode structure for heavy-metal ion capture.
• Built Python pipelines with Scikit-learn and TensorFlow to compare supervised-learning models for multi-component heavy-metal recognition.
• Helped develop a low-cost, non-destructive prototype for meat and soil testing using electrochemical measurements and data-driven analysis.

I also build small, focused tools for AI-native development workflows:

• docs-drift-radar: catch stale README snippets, CLI help, and OpenAPI docs in CI.
• pr-evidence-pack: turn a PR diff into a reviewer-ready evidence pack.
• browser-skill-forge: record browser workflows and export Playwright / agent skills.
• Crysio: a local crystal-structure workbench for visualization, editing, and VASP relaxation-path comparison.
• Auto_PaperFetcher: automate paper fetching and research-workflow collection.
• vscode-remote-ssh-proxy-codex: improve remote SSH / proxy workflows for coding agents.

Programming and engineering

AI and data science

• Graph neural networks, ALIGNN, ML potentials, diffusion models, reinforcement learning, LSTM/SVR pipelines
• Computer vision for sensing and inspection: U-Net style models, MobileViT-style lightweight models
• Scientific data pipelines: feature engineering, active learning, uncertainty/OOD sampling, benchmark evaluation

Scientific computing and simulation

• First-principles and electronic structure: VASP, CP2K, Quantum ESPRESSO, Materials Studio, OpenMX, ABACUS, DeepH
• Molecular dynamics and chemistry simulation: GROMACS, Gaussian, LAMMPS
• Multiphysics and electromagnetic simulation: COMSOL, CST Studio Suite
• Data analysis and research tooling: Origin, SPSS, Python scientific stack, Linux HPC workflows
• Chemical experiments: materials synthesis, annealing, combustion, hydrothermal preparation, electrochemical sensing, electrode-material preparation

These simulation and experimental skills let me contribute to many published materials papers through DFT modeling, RCS / electromagnetic simulation, COMSOL analysis, Python data processing, and mechanism interpretation.

1. G. Zeng, H. Huang, et al. Machine-Learning and Atomic-Scale Mechanistic Insights for Designing Gradient Porous MOF-Derived Carbon Electrodes. Inorganic Chemistry, 2025.
   First-author work combining ML, DFT/MD, and interpretable design rules for MOF-derived carbon electrodes.

2. N. Wang, X. Kou, G. Zeng, et al. Geometry-defect-spin coupling in chiral high-entropy systems: Multiscale mechanisms of GHz electromagnetic dissipation. Science Advances, 2025.
   Multiscale study of chiral high-entropy systems for broadband electromagnetic dissipation.

• Meritorious Winner, The Mathematical Contest in Modeling
• Third Prize, Asia and Pacific Mathematical Contest in Modeling
• Academic Excellence Scholarship, Sichuan Agricultural University
• 20+ awards across national, provincial, and university-level competitions

I want to apply my experience across simulation, experiments, AI4S, and open-source development to more capable research and intelligent systems, especially AutoResearch / AI for Science, embodied intelligence, and LLM development.

• AutoResearch systems that can read papers, call scientific tools, run simulations, and report evidence.
• ML potential and materials-informatics workflows that scale from small datasets to real discovery loops.
• Embodied-intelligence systems that connect perception, planning, and action in real environments.
• LLM applications and agents that are reliable, tool-aware, and useful in complex technical workflows.
• Open-source developer tools that make research and engineering workflows easier to reproduce.

Outside research, I like long-distance cycling and endurance challenges. I once rode solo for 18 days from Chengdu to Lhasa, covering 2,300+ km across many high-altitude mountain passes. That experience shaped how I think about hard problems: break the route down, keep moving, and do not be afraid of difficult terrain.

For the September 2026 domestic postgraduate recommendation process in China, I hope to find a strong research group where I can keep pushing AI4S, scientific simulation, and autonomous materials discovery forward.