Senior Machine Learning Engineer – Physical AI

Goddard•Wilmington, MA

1d•Onsite

About The Position

We are looking for a Senior Machine Learning Engineer to own the AI/ML foundation of our physical AI initiative. This is not a role for someone who builds models in isolation and hands them off — you will be expected to own the full ML lifecycle, from raw sensor data to a model running on constrained hardware in the real world. You will work directly with embedded software, hardware, and systems engineers to bring AI capabilities into physical devices, and you will be accountable for the quality, reliability, and maintainability of every layer you touch. If you take pride in understanding how your model actually behaves on device, have strong opinions about data quality, and hold yourself to a high bar without being told to, you will thrive here.

Requirements

5+ years in machine learning engineering or applied ML, with a demonstrated track record of shipping models to production environments.
Strong proficiency in Python; hands-on experience with PyTorch or TensorFlow for model development and training.
Demonstrated experience optimizing and deploying models to edge or resource constrained targets using TFLite, ONNX, CoreML, TensorRT, or equivalent.
Experience building and maintaining time-series or sensor data pipelines, including preprocessing, feature engineering, and data quality validation.
Working knowledge of quantization, pruning, knowledge distillation, and other techniques for reducing model footprint and inference latency.
Proficiency with experiment tracking tools (MLflow, Weights & Biases, or equivalent), model registries, and automated evaluation and testing workflows.
Solid fundamentals — Git, code review, unit testing, and CI/CD — applied consistently to ML code, not just application code.
Demonstrated ability to work autonomously across hardware and software domains, translate model behavior and limitations clearly to non-ML engineers, and surface risks and uncertainties early.
Working proficiency in C or C++ sufficient to read, review, and meaningfully collaborate on embedded inference integration code; ability to reason about memory layout, execution constraints, and cross-language interface boundaries.

Nice To Haves

Experience with physiological signal processing for medical or wearable applications (ECG, PPG, SpO2, NIBP, IMU, or similar sensor modalities).
Familiarity with FDA guidance on AI/ML-based Software as a Medical Device (SaMD) or practical experience developing software under IEC 62304.
Background in robotics or autonomous systems, including sensor fusion, perception, or closed-loop control.
Experience in a startup or small-team environment where scope, tooling, and process are built alongside the product.

Responsibilities

Design and implement data pipelines for sensor data ingestion, preprocessing, labeling, and curation, ensuring data quality from collection through training.
Train, evaluate, and iterate on ML models for applications including signal processing, anomaly detection, and physiological parameter estimation.
Optimize models for deployment on edge and embedded targets, applying quantization, pruning, and distillation techniques to meet latency and memory constraints.
Deploy models to constrained hardware using TFLite, ONNX, TensorRT, or equivalent runtimes, and validate end-to-end inference behavior on target devices.
Collaborate with embedded software engineers to integrate ML inference into device firmware and software stacks, defining clear interfaces and performance contracts.
Build and maintain MLOps infrastructure: experiment tracking, model versioning, automated evaluation pipelines, and CI/CD for models.
Work with hardware and systems teams on sensor selection, data collection protocol design, and validation methodology.
Document model development, training procedures, validation results, and known limitations to support regulatory submissions and internal quality systems.
Design and execute rigorous model validation: statistical test set design, distributional shift analysis, out-of-distribution detection, and confidence calibration, particularly for safety-relevant outputs.
Proactively identify data quality gaps, model failure modes, and deployment blockers before they reach production.