Hello!

I am Shengqi, a senior undergraduate in Mathematics and Computer Science at University of Wisconsin-Madison (UW-Madison) working with Prof. Frederic Sala's Sprocket Lab and Prof. Benedek Valko.

My research interests are in data-centric AI, foundation models, Mathematics, and trustworthy ML. In particular, I work on:

• Programmatic Evaluation: Developing program-as-a-judge methods that distill LLM judging logic into executable Python functions, then aggregate noisy program outputs through weak supervision to produce reliable preference labels with low cost and high throughput.
• Random Matrix Identities: Studying two-point correlation functions of the $\mathrm{Sine}_\beta$ process through ODEs and symbolic computation. The project develops a framework for deriving explicit correlation formulas by reducing, simplifying, and solving the ODE systems associated with classical beta ensembles. In the $\beta = 6$ case, I independently discovered a factorization that turns a complex third-order ODE into a solvable decomposition of lower-order ODEs, leading to an explicit candidate formula for the pair correlation function.
• Robustness Verification: Developing tools for certifying robustness of Transformer under discrete text perturbations. I study how to extend fast certification methods from convolutional text classifiers to attention-based architectures, and deriving spectral-norm Jacobian bounds for the self-attention operator.