Jihoon Kim
I’m an AI Researcher at Narnia Labs and a Ph.D. student in Mechanical Engineering at KAIST, advised by Prof. Namwoo Kang at Smart Design Lab.
My research makes generative design manufacturable—embedding real-world manufacturing constraints into AI-driven design, so optimized components can actually be mass-produced.
news
| Jun 16, 2026 | PyTopo3D is now archived on Zenodo with a citable DOI (10.5281/zenodo.20697949) — the framework can now be cited directly in publications. |
|---|---|
| Jan 15, 2026 | Published an interactive explainer for Deep Generative Design for Mass Production, where you can explore the generated, manufacturable designs in 3D. Read it → |
| Apr 8, 2025 | Open-sourced PyTopo3D, a pure-Python framework for 3D SIMP topology optimization (pip install pytopo3d), with a paper on arXiv. A modern alternative to MATLAB-based 3D topology optimization. |
| Jan 2, 2025 | Joined Narnia Labs as an AI Researcher, working on generative design and design-for-manufacturing for automotive clients. |
| Aug 31, 2024 | Our paper on deep generative synthesis of four-bar linkage mechanisms was published in the Journal of Computational Design and Engineering (with Sumin Lee and Namwoo Kang). Paper · Code |
research
Generative design has revolutionized how engineers explore design possibilities, yet a fundamental gap remains: most generated designs cannot be manufactured at scale. The intricate geometries from topology optimization and deep generative models often require additive manufacturing—poorly suited for mass production.
My research bridges this gap by embedding manufacturing constraints directly into the generative process. A key insight is that complex 3D manufacturing constraints can be elegantly captured through 2D depth representations, making AI-based design both computationally tractable and aligned with processes like die casting and injection molding.
selected publications
-
Deep Generative Design for Mass Production2024AI-generated designs often can’t be mass-produced. This work converts 3D design problems into 2D depth images, embedding manufacturing constraints directly into the generative process—producing designs ready for die casting or injection molding.
