Hello, I'm
Lei Ge (雷鸽)
PhD Candidate in Design Engineering
Imperial College London
I am a PhD candidate at Imperial College London, expecting to graduate in 2026, and a part-time LLM Machine Learning Engineer at Polaron. My research focuses on LLM interpretability and agentic systems. Before my PhD, I studied fashion design and engineering and researched AI for garment structure design. That engineering background continues to shape how I think about computation, geometry, and the physical world.
News
LLM surrogate belief under sparse observations
I study how language models behave as surrogate models in low-data optimization. The work shows that prompt design and query protocol are core components of surrogate specification, shaping uncertainty, belief updating, and downstream scientific decisions.
LLM-guided optimization for battery charging protocols
I develop LLM agents for battery fast-charging optimization. The agents expand protocol search beyond fixed human parameterizations, incorporate natural-language safety constraints, and operate in expensive, slow-feedback scientific settings.
Knowledge representation in large language models
I investigate how LLMs encode structured scientific knowledge. This work identifies a 3D spiral structure in hidden states that aligns with the conceptual organization of the periodic table.
- P1
Elicitation Matters: How Prompts and Query Protocols Shape LLM Surrogates under Sparse Observations
Lei Ge, Samuel J. Cooper
arXiv:2605.04764 - P2
From Prompt to Protocol: Fast Charging Batteries with Large Language Models
Lei Ge, Ferran B. Planella, Stephen G. Baird, Samuel J. Cooper
arXiv:2601.09626 - P3
Do Llamas Understand the Periodic Table?
Lei Ge, Samuel J. Cooper
Digital Discovery, 4(12), 3455–3465 - P4
Materials Science in the Era of Large Language Models: A Perspective
Lei Ge, Ronan Docherty, Samuel J. Cooper
Digital Discovery, 3(7), 1257–1272
Computational fashion
Before my PhD, I worked across digital pattern-making, computer vision, personalized apparel, and zero-waste design. That training gives my AI research a distinct perspective: systems should be interpretable, materially grounded, and attentive to constraints.
View fashion projects