CSE Community Seminar | February 27, 2026

Abstract

Generative AI is a technological marvel that has yet to transform engineering design at scale. In my talk, I argue that bridging this gap requires three missing ingredients: precision (reliable constraint satisfaction), integration (alignment with real design workflows), and generality (transferability across domains). First, I discuss the counterintuitive juxtaposition of probabilistic models in high-stakes engineering settings. I propose new approaches to more reliably formulate, evaluate, and train generative models, such as leveraging counterexamples to learn design constraints. Second, I illustrate how AI’s disconnect from design practice limits its adoption. I present more grounded benchmarks and new computational design tools that forego end-to-end design in favor of small, surgical design modifications. Third, I showcase synthetic-data pipelines that allow models to acquire reusable engineering knowledge, enabling a single model to solve disparate engineering problems across multiple domains. Together, these advances lay the foundation for AI systems that are as reliable and versatile as the engineers who use them.