Research

Agentic AI for Autonomous Design Optimization

Research Overview

My research automates Design Optimization,
spanning Generative, Predictive, and Optimization AI for engineering design
and now building LLM agents that autonomously perform decomposition-based Design Optimization.

1
Generative AI
Creating design candidates
  • DialogCAD
  • DeepJEB (co-author)
2
Predictive AI
Replacing simulation with neural surrogates
  • Point-DeepONet
  • BMO-GNN
  • Physics-Constrained GNN (co-author)
3
Optimization AI
Optimizing & orchestrating with LLM agents
  • AutoATC
  • LLM-Guided Penalty Parameter Updates
Ph.D. Dissertation — Agentic AI for Autonomous Analytical Target Cascading
LLM agents that autonomously decompose and optimize engineering systems,
from academic benchmarks to industrial CAD assemblies

Applied with Hyundai Motor · Samsung Heavy Industries · LG Display · KORAIL

Proposed Frameworks

Generative AI
DialogCAD

A dialogue-driven CAD generation framework powered by a multi-agent system over the Model Context Protocol.

DeepJEB (co-author)

A synthetic 3D dataset of jet engine brackets generated by deep learning for data-driven design.

Predictive AI
Point-DeepONet

Predicts nonlinear physical fields on arbitrary 3D shapes under varying load conditions.

BMO-GNN

Bayesian mesh optimization that boosts engineering performance prediction with graph neural networks.

Physics-Constrained GNN (co-author)

Spatio-temporal prediction of drop impact on OLED display panels with physics constraints.

Optimization AI
AutoATC

Automatically decomposes complex multidisciplinary design problems using LLM-based agents.

LLM-Guided Penalty Parameter Updates

An LLM adaptively tunes penalty parameters inside Analytical Target Cascading for constrained design.