Rockwell Automation
SLM Training Engineer
Milwaukee, WI and 3 more · On-site · Unspecified
Sites: Milwaukee, WI, Austin, TX, Houston, TX, Mayfield Heights, OH
- Discipline
- Other
- Industry
- Manufacturing
Employer description
Rockwell Automation is a global technology leader focused on helping the world’s manufacturers be more productive, sustainable, and agile. With more than 28,000 employees who make the world better every day, we know we have something special. Behind our customers - amazing companies that help feed the world, provide life-saving medicine on a global scale, and focus on clean water and green mobility - our people are energized problem solvers that take pride in how the work we do changes the world for the better.
We welcome all makers, forward thinkers, and problem solvers who are looking for a place to do their best work. And if that’s you we would love to have you join us!
Job Description
Job Description
Rockwell Automation is looking for an experienced SLM and Decision Model Training Engineer to build the small, fast, specialized models that make our enterprise AI agents reliable and affordable at scale.
This role sits within our AI Engineering organization and focuses on two model families: small language models (SLMs) fine-tuned for Rockwell domains and tasks, and decision models that return typed, calibrated judgments (choices, scores, and yes/no probabilities) instead of free-form text, in the style of “System One” models such as TypeSafe’s Jev. These models will route requests, classify and triage work, act as guardrails on agent actions, and evaluate agent outputs, so that large language models are used only where open-ended reasoning is truly needed.
You have a proven track record of training, fine-tuning, evaluating, and deploying machine learning models in production, with a strong grounding in probability calibration and model evaluation.
This is a Hybrid role. You will report to the Director of Agentic AI.
Your Responsibilities
Train, fine-tune, and distill small language models for Rockwell-specific domains and agent tasks, such as tool selection, structured extraction, summarization, and industrial and product terminology. Build decision models that take text or structured state plus developer-defined questions and return typed choices, ordered scores, or yes/no probabilities that software can act on directly. Train decision models to produce well-calibrated probabilities, so confidence scores can drive thresholds for automating confident cases and escalating uncertain ones to people. Apply decision models across the agent platform for model routing, request triage, guardrails (for example, whether an agent should make a tool call), and automated evaluation of LLM and agent outputs. Build data pipelines for training and evaluation, including data curation, labeling strategies, synthetic data generation, and distillation from larger teacher models. Design evaluation suites that measure accuracy, calibration, latency, and cost against LLM baselines, and run model-versus-model comparisons to guide when to use an SLM, a decision model, a classical classifier, or a large model. Optimize models for production and edge deployment through quantization, pruning, and efficient serving, including scenarios close to industrial and plant-floor environments. Operate the model lifecycle with MLOps practices: experiment tracking, model registry, version pinning, monitoring for drift and calibration decay, and retraining. Partner with agent, platform, and business teams across the enterprise to identify high-volume, bounded decisions that benefit from specialized models, and to make build vs. buy recommendations against commercial decision and small-model offerings.
Key Technical Skill Expectations
Small Language Model Training: Fine-tune and adapt open-weight SLMs (e.g., Phi, Llama, Mistral, Qwen, Gemma) using supervised fine-tuning, parameter-efficient methods such as LoRA and QLoRA, and preference optimization (e.g., DPO), with frameworks such as PyTorch, Hugging Face Transformers, TRL, and DeepSpeed.
Decision & Discriminative Models: Build transformer-based classifiers, rankers, and scoring models (e.g., encoder models and cross-encoders) that output typed decisions over a fixed answer space, including multi-label, ordinal, and zero-shot setups where categories are supplied at inference time.
Calibration & Uncertainty: Apply probability calibration methods (e.g., temperature scaling, isotonic regression) and reward-based training for calibrated decisions, and measure calibration with reliability diagrams, expected calibration error, and Brier score.
Distillation & Data Engineering: Generate and curate training data, including LLM-labeled and synthetic data, apply knowledge distillation from larger teacher models, and manage data quality, bias, and label noise.
Evaluation & Benchmarking: Build offline and online evaluations that compare SLMs, decision models, and LLMs on accuracy, calibration, latency, and cost, and use LLM-as-judge and decision-model-as-judge patterns responsibly.
Model Optimization & Serving: Quantize and compress models (e.g., GGUF, ONNX, INT8/INT4), and serve them efficiently with tools such as vLLM, ONNX Runtime, Triton Inference Server, or Azure AI Foundry, including edge and on-premises deployment.
MLOps & Governance: Use MLflow or Azure Machine Learning for experiment tracking, model registry, CI/CD for models, monitoring, and responsible AI practices that comply with enterprise governance and security policies.
The Essentials - You Will Have
Bachelor’s Degree in Computer Science, Machine Learning, Data Science, Statistics, or a related field. Legal authorization to work in the U.S. We will not sponsor individuals for employment visas, now or in the future, for this job opening.
The Preferred - You Might Also Have
Typically requires a minimum of 5 years of relevant professional experience, including 3+ years training and deploying machine learning or NLP models in production. Master’s degree or PhD in Computer Science, Machine Learning, Statistics, or a similar discipline. Hands-on experience fine-tuning, distilling, and deploying small language models for specific enterprise tasks. Familiarity with System One decision models such as TypeSafe’s Jev, and with Reinforcement Learning for Calibrated Decisions (RLCD) or similar approaches that reward calibrated probabilities. Experience with reinforcement learning methods for language models (e.g., RLHF, RLAIF, GRPO) and reward modeling. Experience building model routing, guardrail, or LLM evaluation systems for agentic AI applications. Experience deploying models to edge or industrial environments, or working with industrial and manufacturing data. Proficiency in Python and PyTorch, and experience with GPU training infrastructure and distributed training. 3+ years of experience with major cloud platforms, such as Microsoft Azure, AWS, or Google Cloud Platform (GCP). Experience working and coordinating with a geographically diverse team across multiple time zones. Familiarity with Scaled Agile Framework (SAFe) methodologies.
What We Offer
Health Insurance including Medical, Dental and Vision 401k Paid Time off Parental and Caregiver Leave Flexible Work Schedule where you will work with your manager to enjoy a work schedule that can be flexible with your personal life To learn more about our benefits package, please visit at www.raquickfind.com We are an Equal Opportunity Employer including disability and veterans.
This position is part of a job family. Experience will be the determining factor for position level and compensation.
#LI-Hybrid
#LIfeAtRok
#LI-SB1
We are an Equal Opportunity Employer including disability and veterans.
If you are an individual with a disability and you need assistance or a reasonable accommodation during the application process, please contact our services team at +1 (844) 404-7247.
Rockwell Automation’s hybrid policy aligns that employees are expected to work at a Rockwell location at least Mondays, Tuesdays, and Thursdays unless they have a business obligation out of the office.
Stateside Spec lists roles for engineering teams in the United States. Work-authorization notes come from the employer. We do not guarantee visa policy. Apply off-site. Report an issue with this listing View employer source