Root Access

Machine Learning Engineer

Reposted One Month Ago

New York City, NY, USA

In-Office

Mid level

Artificial Intelligence • Software • Design

Rage With the Machines

The Role

Design and train physics-informed deep learning models (PINNs, FNOs, Neural Operators) for electromagnetic and heat PDEs; build ECAD-to-tensor/graph data pipelines; calibrate models with lab measurements (VNA, TDR, EMI); integrate multimodal upstream models (GNNs/LLMs); and optimize GPU training/inference for real-time (<100ms) execution.

Summary Generated by Built In

About the company
Root Access is a frontier electronics company. We are a NYC-based startup funded by top investors. Our team is a passionate mix of engineers across electrical, firmware, software, and machine learning.

Core Responsibilities

Architect Physics Foundation Models: Design and train deep learning models.
Build the ECAD Data Pipeline: Develop high-performance asset pipelines to convert geometric, discrete, and multi-layer PCB files (ODB++, IPC-2581, STEP, Gerber) into continuous space data.
Multi-Modal Architecture Integration: Collaborate on connecting upstream Graph Neural Networks (GNNs) or LLMs mapping schematic topologies to downstream spatial physics engines.
Optimize for Real-Time Execution: Optimize training and inference pipelines on GPU clusters.

Required Technical Skills & Qualifications

Education: Master’s or Ph.D. in Computer Science, Mathematics, EE, Physics, or a related quantitative field with a focus on Scientific Machine Learning (SciML).
Deep Learning Frameworks: 4+ years of expert-level experience with PyTorch or JAX.
SciML Expertise: Direct, hands-on experience building and training PINNs, FNOs, etc.
Mathematical Depth: Exceptional understanding of partial differential equations (PDEs), vector calculus, automatic differentiation (autograd), and numerical optimization algorithms (Adam, L-BFGS).
Data Pipelines: Strong proficiency in manipulating spatial or geometric datasets using Python libraries (NumPy, SciPy, Shapely, Open3D, or custom voxelization matrices).

Skills Required

Master's or Ph.D. in Computer Science, Mathematics, Electrical Engineering, Physics, or related quantitative field with focus on Scientific Machine Learning (SciML).
4+ years of expert-level experience with PyTorch or JAX.
Direct hands-on experience building and training PINNs, DeepONets, or Fourier Neural Operators (FNOs).
Experience using SciML frameworks such as NVIDIA Modulus, DeepXDE, or PyTorch Geometric.
Exceptional understanding of PDEs, vector calculus, automatic differentiation (autograd), and numerical optimization algorithms (Adam, L-BFGS).
Strong proficiency manipulating spatial/geometric datasets using NumPy, SciPy, Shapely, Open3D, or custom voxelization matrices.
Experience converting ECAD formats (ODB++, IPC-2581, STEP, Gerber) into continuous tensor grids, signed distance fields (SDFs), or graph embeddings.
Experience optimizing training and inference pipelines on GPU clusters for sub-100 millisecond forward-pass performance.
Familiarity integrating multimodal architectures (Graph Neural Networks or LLMs) with spatial physics engines.