Data Scientist

Build risk-scoring models over synthetic tabular data, engineering features from curated medallion-layer tables. Build anomaly and outlier detection to surface irregularities in records and process data. Build optimization models for prioritization, routing, and resource allocation. Validate honestly - calibration, discrimination, stability, explainability. A correctly characterized model matters more than a flattering headline metric. Package deliverables as jobs and Asset Bundles, tracked in MLflow, and document assumptions, limitations, and what must be revalidated against real data post-ATO.

U.S. citizenship and active T5/SSBI federally adjudicated clearance required. Hands-on Databricks. Feature engineering on tabular and time-series data - encoding, aggregation, leakage prevention, and selection grounded in domain reasoning rather than automated search alone. Supervised learning on tabular data: gradient boosting (XGBoost/LightGBM), regularized regression, and the judgment to know when the simpler model is the right answer. Model calibration and evaluation under class imbalance - you can explain why AUC alone is insufficient for a risk score. Anomaly detection: isolation forests, autoencoders, statistical process control, or comparable - with a clear account of how you validated detections without labels. Optimization: LP/MIP or heuristic methods (OR-Tools, Pyomo, SciPy, or equivalent) applied to a real allocation or prioritization problem. Explainability (SHAP or comparable) in a decision-support context. Privacy-preserving synthetic data generation from CUI, PII, or comparably restricted source data - relational tabular data with distributional fidelity, cross-column correlations, referential integrity, and preservation of the rare-event structure that anomaly detection and risk scoring depend on. Includes an understanding of re-identification risk. Strong Python, SQL, and Spark. Government or defense contracting experience., Modeling on federal investigative, vetting, fraud, or insider-threat data. Direct experience with FedRAMP, NIST 800-171, CMMC L2, or CUI handling. Familiarity with LLM/GenAI workflows - useful for collaboration with a peer document-intelligence workstream, but secondary to the core ML skill set. H2O (Driverless AI, H2O-3). MLflow, Databricks Asset Bundles, Unity Catalog. Fairness / adverse-impact analysis in a regulated or decision-support setting.

Soft Skills Self-directed execution against a fixed milestone with minimal oversight. Honest reporting of model behavior - comfortable stating what synthetic-data performance does and does not establish about real-world accuracy. Collaboration across technical and non-technical teams. Clear documentation and active knowledge transfer.