> Markdown version of [/jobs/ext/3786193-ml-engineer-work-from-home](https://www.wearedevelopers.com/jobs/ext/3786193-ml-engineer-work-from-home). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md). --- # ML Engineer - Work From Home - **Company:** TRENT HOEK OUTDOORS LLC - **Location:** United States (Remote available) - **Experience:** Experienced - **Salary:** $90,000.0 - $120,000.0 - **Contract:** Permanent contract - **Skills:** Information Leak Prevention, Database Queries, Information Extraction, Information Theory, Python (Programming Language), Machine Learning, Language Modeling, NumPy, Performance Tuning, Raw Data, Parquet, Supervised Learning, Feature Engineering, Pytorch, SHAP (Shapley Additive Explanations), Apache Spark, Qwen, Deep Learning, Pandas, Scikit Learn, Information Technology, Xgboost, Free and Open-Source Software, Model Quantization, Machine Learning Operations, Feature Extraction, Evaluation of Large Language Models, Data Pipelines - **Published:** October 16, 2026 - **Apply:** https://arc.dev/remote-jobs/j/redirect/przakq7c5u ## About the Role *At least 3 years of experience training and deploying machine learning models, including end-to-end ownership of a system from raw data through validated results. *Hands-on experience training transformer architectures in PyTorch, including at least one self-supervised or contrastive learning objective on non-text data. *Experience deploying open-weight language models for structured information extraction and measuring accuracy against a labeled reference dataset. *Demonstrated ability to handle model calibration, class imbalance, and leakage prevention using real-world observational data. *Experience with gradient-boosted learning methods such as LightGBM, XGBoost, or CatBoost and SHAP-based explainability. *Advanced Python skills with NumPy, pandas, scikit-learn, and PyTorch. *Strong SQL skills and experience using Spark for large-scale feature engineering. *Applied mathematical knowledge covering probability, statistical inference, linear algebra, optimization, information theory, evaluation metrics, and causal reasoning. *Ability to work within access-restricted environments using approved tools and CPU-only computing. *U.S. work authorization and the ability to obtain a Public Trust or higher background determination. Preferred Qualifications *Master's degree in mathematics, statistics, or applied mathematics with a concentration in data science. Computer science or another quantitative degree is also considered when supported by equivalent mathematical depth in academic coursework and professional experience. *Published research or open-source contributions involving tabular deep learning, event-sequence modeling, or LLM evaluation. *Experience with model quantization and large-scale CPU inference optimization. *Familiarity with healthcare claims or administrative datasets, including diagnosis and procedure coding systems. *Experience with fairness assessment and subgroup evaluation. *Experience working in FedRAMP, HIPAA, or other access-restricted environments. *Active or previous federal Public Trust determination. Technical Stack *Representation Learning: SAINT, TabTransformer, FT-Transformer, masked-feature reconstruction, and contrastive learning. *Temporal Modeling: Transformer encoders for ordered event sequences, including BEHRT-style approaches, and temporal convolutional networks. *Language Models: Locally hosted Llama, Mistral, and Qwen model families; llama.cpp or equivalent CPU inference tools. *Supervised Learning: LightGBM, XGBoost, CatBoost, and scikit-learn. *Deep Learning: PyTorch. *Explainability: SHAP, TreeSHAP, KernelSHAP, and integrated gradients. *Data and MLOps: Python, Spark, Parquet, pandas, NumPy, and MLflow. ## Description iAdeptive Technologies is seeking a Machine Learning Engineer to lead model development for a federal data and AI program. The role focuses on self-supervised representation learning, temporal modeling, language-model-based information extraction, supervised classification, model calibration, and explainability. You'll take ownership of the modeling lifecycle, from architecture selection and experimentation to training, validation, and reproducible results. The work involves large, linked administrative datasets in a restricted analytical environment with approved tools, no external connectivity, and CPU-only computing. This is a modeling-focused position rather than a data pipeline role. Strong mathematical reasoning and the ability to evaluate models critically are essential. Key Responsibilities *Design and train self-supervised tabular transformers, including SAINT, TabTransformer, and FT-Transformer, along with temporal sequence models for longitudinal event data. *Select pretraining objectives, run experiments, evaluate embedding quality, and benchmark architectures against gradient-boosted baselines. *Deploy locally hosted, open-weight language models to extract structured information from unstructured records using prompt design, constrained output schemas, and post-processing. *Optimize language models for high-volume CPU inference through quantization and performance tuning while maintaining acceptable extraction accuracy. *Implement dual-coding evaluation protocols using stratified samples, independent human coding, and field-level agreement measurements before large-scale extraction. *Train gradient-boosted classifiers using LightGBM, XGBoost, or CatBoost on learned embeddings, accounting for severe class imbalance. *Calibrate classification thresholds based on the expected cost of different error types and report calibration alongside discrimination metrics. *Identify and prevent data leakage across temporal splits and linked record sources. *Generate SHAP-based feature attributions at the individual-prediction and cohort levels, assess subgroup patterns, and verify attribution stability across model retraining. *Maintain supporting feature and inference workflows using Python, Spark, and Parquet. *Use MLflow to track experiments and version datasets and models so reported results can be reproduced. *Document model architecture decisions, experimental findings, validation results, and the reasoning behind technical choices.