Machine Learning Engineer - Physical AI in San Diego, California at Koh Young America, Inc.

Koh Young Technology

Koh Young Technology, founded in 2002 in Seoul, South Korea, is the world leader in 3D measurement and inspection technology used in the production of micro-electronics assemblies. Using patented 3D technology, Koh Young provides best-in-class products in Solder Paste Inspection(SPI) and Automated Optical Inspection(AOI) for electronics manufacturers worldwide.

In addition to our headquarters in Seoul, Korea, we maintain offices in Europe (Germany), United States, Japan, Singapore and China, allowing us to have close communication with our customers and have global network of sales and services.

The company invests very heavily in R&D (60% of the workforce, >10% of sales) and treasures talented team members. Koh Young has a dynamic innovative culture and strives to be at the top of the industry.

Koh Young Research America (KYA)

Located in San Diego, CA, Koh Young America has been serving business partners in North America with sales and technical support since 2010. In addition to KYA’s sales and support for North America, Koh Young Research America was established in 2016 in San Diego, CA as a Research & Development hub for artificial intelligence and deep learning technologies.

Summary

Manufacturing is entering a new era — one where AI doesn't just learn from data, but understands the physics behind every process. At Koh Young, the global leader in 3D inspection technology, we're building intelligent systems that predict, optimize, and control manufacturing processes from first principles.

This role is about building custom neural architectures from scratch for physical systems — not fine-tuning foundation models or assembling LLM pipelines.

We're looking for a Process ML Engineer who lives at the intersection of deep learning, physics, and statistics. Someone who can design state-of-the-art AI architectures — transformers, diffusion models, neural operators — that respect process causality, formulate a PDE for a new domain, and take the model all the way to production hardware.

The models you build will run inside inspection machines deployed on thousands of production lines worldwide, making decisions on millions of components every day. You'll have direct access to real process data, domain experts, and the hardware your models ship on.

Responsibilities

• Architect neural networks with physical inductive bias — encoding process causality and sequential dynamics directly into network structure, not just loss functions
• Design multi-modal architectures that fuse heterogeneous signals (geometry, process parameters, temporal sequences, sensor data) into unified learned representations
• Develop physics-informed and data-free models (PINNs, neural operators, differentiable simulators) for process prediction, optimization, and diagnostics
• Formulate governing equations from first principles (fluid dynamics, thermodynamics, rheology, contact mechanics, etc.) and embed them into learnable representations
• Apply Bayesian optimization, surrogate modeling, Gaussian processes, and DoE for high-dimensional process parameter optimization
• Analyze 3D metrology and inspection data, define quality metrics (Cpk, SPC), and close the loop between measurement and process control
• Optimize models for low-latency, on-device inference in production equipment
• Collaborate across physics, software, and manufacturing teams to deploy models that run autonomously on the factory floor

Skills and Qualifications

• M.S. or Ph.D. in Mechanical Engineering, Physics, Computer Science, Electrical Engineering, or a related quantitative field, with 4+ years of relevant experience
• Deep fluency in PyTorch; comfortable writing custom architectures, autograd functions, and training loops from scratch
• Strong understanding of multi-modal learning — the ability to design embeddings that unify diverse physical domains (geometry, dynamics, process conditions) into a shared representation space
• Solid grounding in continuum mechanics: fluid dynamics, thermodynamics, rheology, heat transfer, solid/structural mechanics, etc. — enough to formulate and simplify PDEs for real engineering processes
• Hands-on implementation experience with at least one of: PINNs, Fourier Neural Operator, DeepONet, differentiable physics, or causality-aware architectures
• Strong statistical foundations: Bayesian inference, Gaussian processes, uncertainty quantification
• Track record of deploying models into production systems
• Comfort with ambiguity: you'll often start with a physical process, not a dataset

Have at least one of the followings:

• Experience building end-to-end ML pipelines — from problem formulation and data generation to training, evaluation, and deployment
• Demonstrated ability to translate domain knowledge into model architecture or feature design decisions
• Hands-on experience in manufacturing processes — semiconductor, SMT, electronics assembly, or precision manufacturing — with understanding of how process variables affect product quality
• Published work or projects in scientific machine learning, computational physics, or neural operators
• Experience with SPC, Cpk analysis, or 3D inspection/metrology systems
• Familiarity with physics-ML platforms or digital twin frameworks

Benefits

• Health/Dental/Vision/Life Insurance at NO employee premium (including dependent coverage)
• 401(k) retirement plan (Immediately 100% vested)
• Generous PTO and paid holidays