About me

Hi, I’m Dr. Ziqi Guo, a Machine Learning Researcher at Apple. I use AI to make physics simulation faster and engineering design smarter. My applications span batteries (predicting the performance of new cell designs), thermal management and heat transfer (phonon and thermal-conductivity simulation), radiative cooling and optics (nanoparticle paints and infrared properties), and semiconductor and electronic materials (hBN, β-Ga2O3, and Al/Si interfaces). The methods behind them include fine-tuning transformer-based and atomistic foundation models, physics-informed neural networks, graph and recurrent neural network surrogates, generative inverse design with multi-objective optimization, and GPU-accelerated scientific computing. I also build agentic AI systems, with LLM agents and agent memory (hybrid retrieval, knowledge graphs), that automate the manual engineering workflows around them.

I received my Ph.D. in Mechanical Engineering from Purdue University (2025), co-advised by Prof. Xiulin Ruan and Prof. Guang Lin. Before that, I earned my B.E. with Honors from Huazhong University of Science and Technology.

What I work on

AI for scientific simulation and design

• Foundation models for battery design. At Apple, I use a fine-tuned foundation model with physics priors in place of empirical rules to understand battery design and physics, cutting prediction error by 30% and predicting new cell designs before hardware is built.
• Accelerating thermal and optical simulation. I combined machine learning, statistical sampling, and GPU computing to speed up first-principles thermal transport (phonon) calculations and photon Monte Carlo optical simulations by 100× and more, and released the results as open-source codes: FourPhonon for thermal transport and FOS for optical spectra.
• Generating new material designs. I built surrogate-driven generative inverse design to find new material formulations, for example colored radiative cooling paints with nearly double the visible reflectance at matched color.

Atomistic-scale simulation and foundation models

• I study materials at the atomistic scale with first-principles simulation and the AI-for-materials paradigm, including fine-tuning pretrained atomistic foundation models (universal ML force fields) to get ab initio accuracy at a fraction of the cost.

Agentic AI

• LLM agents for engineering workflows. I build LLM agents that fully automate performance prediction for new product delivery, cutting a one-week manual workflow to one day.
• Agent memory systems. I build agent memory systems for team applications: hybrid retrieval (RAG), knowledge graphs, and persistent shared memory that let agents recall, update, and reuse team knowledge across tasks.

In total I have published 20+ papers (see Publications or my Google Scholar profile), with 560+ citations and an h-index of 12. My work has been recognized with several awards and honors.