CAD Automation Researcher

San Francisco | 5 Days Onsite

Location:

Onsite in San Francisco

Compensation:
Competitive Salary + Equity

Geophysics + Seismic Researcher

Engineering simulation is one of the last major categories of software that AI hasn't rebuilt. The tools used to design aircraft, ships, reservoirs, and medical devices still run on numerical methods that are decades old, and an engineer can wait a full day for a single answer. UniversalAGI is building foundation models that learn physics directly from data, and they are already running in early deployments on real computational fluid dynamics and reservoir engineering problems for some of the largest industrial and defense organizations in the world.

We are a team of 25 researchers and engineers in San Francisco backed by Elad Gil (#1 Solo VC), Eric Schmidt (former Google CEO), Prith Banerjee (ANSYS CTO), Ion Stoica (Databricks Founder), Jared Kushner (former Senior Advisor to the President), David Patterson (Turing Award Winner), and Luis Videgaray (former Foreign and Finance Minister of Mexico).

As a CAD Automation Researcher you'll be in the arena from day one, building the future of AI for physics simulation with your own hands. This is your chance to take everything you've learned about CAD geometry, mesh generation, and design optimization and use it to build foundation AI models that remove the single biggest bottleneck in engineering simulation today: turning messy, real world CAD into something a computer can actually reason about.

You'll work directly with the CEO and founding team to shape our product around the real pain points you've faced in your career: the hours lost repairing broken CAD files, the manual de-featuring before a single simulation can even run, and the trial and error of hand tuned topology optimization. You'll build the models that make that friction disappear.

• Develop methods to automatically repair, defeature, and simplify raw CAD geometry into simulation ready form, removing today's largest manual bottleneck in our design pipeline.

• Develop general purpose optimization techniques, including topology optimization, shape optimization, and differentiable design search, that operate directly on learned geometry representations and AI physics model outputs.

• Design and train models, such as implicit neural representations, graph neural networks, and diffusion models, for geometry generation and reconstruction from imperfect, incomplete, or multi format CAD inputs (STEP, IGES, mesh, point cloud) or embeddings.

• Own your workflow end to end: synthetic geometry data generation, training, evaluation, and deployment into UniversalAGI's production simulation pipeline.

• Collaborate directly with the CEO, founding team, and our physics research group to make sure geometry and optimization models integrate cleanly with the rest of the foundation model stack.

• Help define UniversalAGI's technical point of view on AI driven geometry generation and design optimization, and publish results internally and, where appropriate, externally.

• PhD in Computer Science, Mechanical Engineering, Computational Geometry, or a related field, or equivalent hands on industry research experience.

• 2+ years building deep learning models for geometry processing, computer graphics, computational geometry, or 3D generative modeling.

• Deep, hands on understanding of CAD representations and geometry processing: boundary representations (B-rep), CAD kernels such as Open CASCADE or Parasolid, mesh generation, and defeaturing or healing techniques.

• Proven research track record applying modern architectures, such as graph neural networks, implicit neural representations, transformers, or diffusion models, to 3D geometry or physical design problems.

• Practical experience with optimization methods relevant to physical design, such as topology optimization, shape optimization, or differentiable and gradient based design search.

• Expert Python and deep learning framework proficiency (PyTorch or JAX), with the ability to build and train models from scratch rather than only fine tune existing ones.

• High execution velocity and comfort operating with significant ownership in a fast moving startup environment.

• Published research in top tier venues such as NeurIPS, ICML, ICLR, SIGGRAPH, or CVPR.

• Experience building on or extending commercial CAD kernels or CAD or CAM software internals.

• Familiarity with reinforcement learning or large language model based…