Machine Learning Engineer

Compensation: $45.67/hr to $57.69/hr. Exact compensation may vary based on several factors, including skills, experience, and

. Benefit packages for this role will start on the 31st day of employment and include medical, dental, and vision insurance, as well as HSA, FSA, and DCFSA account options, and 401k retirement account access with employer matching. Employees in this role are also entitled to paid sick leave and/or other paid time off as provided by applicable law. Day-to-day You will work as an ML systems engineer building software for non-destructive testing in the nuclear industry (e.g., visual, ultrasonic, eddy current).

will be to support the development of the ML workflows by ensuring that they can be deployed in reproducible, stable workflows for pilot use. You’ll collaborate with subject-matter experts to validate solutions and follow best practices for packaging, observability, and integration into existing systems. Early projects will have well-defined scopes focused on technical execution, with more flexibility and responsibility for shaping deployment patterns as the team grows. Prerequisites • Bachelor’s degree in Computer Science, Physics, Software Engineering, Applied Mathematics, Data Science, or equivalent degrees with the appropriate technical background; advanced degrees (MS/PhD) welcome but not expected • 2–5 years of professional software engineering experience, with at least 1 of those years in a hands-on role in a machine learning engineering or data science context • Demonstrated experience packaging, deploying, or supporting data-driven or ML-based Python systems beyond exploratory notebooks across Windows and Linux environments • Comfortable working in on-prem, air-gapped, regulated, or otherwise constrained, systems Preferred skills • Build data adapters or ETL workflows to convert legacy scientific/industrial formats into ML‑ready datasets; support local artifact and dataset management in constrained environments. • Embed core ML logic into reproducible training and inference pipelines, ensuring clean separation of configuration, integration code, and runtime concerns. • Package and deploy ML code for stable execution in constrained or air‑gapped environments, managing dependencies, environment isolation, and offline realities, with deployment targets ranging from edge devices to HPC environments. • Integrate ML workflows with legacy applications by capturing runtime constraints, hardware considerations, and deployment limitations. • Implement structured logging, error handling, and performance tracing to support debugging and diagnostics without overengineering. • Configure and use on-prem object stores, data/versioning tools, and model synchronization workflows. Early Growth (Expected after 3–6 months) • Begins standardizing deployment and reproducibility practices across multiple projects to shorten setup time for future work. • Contributes informed input into architectural decisions related to artifact storage, configuration patterns, and pipeline structure in collaboration with data scientists. • Identifies recurring integration challenges with legacy systems and proposes pragmatic improvements within existing ownership boundaries. • Develops stronger intuition for failure modes in pilot deployments and proactively mitigates them before field testing. Long-Term Development • Takes ownership of reproducibility and deployment standards across multiple projects. Moves from stabilizing individual workflows to defining consistent conventions for configuration management, artifact handling, and pilot deployment that reduce setup time and ambiguity for future projects. • Begins shaping integration boundaries in collaboration with Applied ML Engineers and legacy system owners. Contributes to defining stable training and inference interfaces, data contracts, and runtime assumptions that reduce coupling and minimize future rewrite risk. • Proactively identifies and mitigates recurring failure modes in pilot environments. • Develops intuition for hardware variability, data inconsistencies, runtime edge cases, and integration friction, and introduces pragmatic safeguards and diagnostics before issues surface in field testing.