Data Scientist

Bigbear.ai•McLean, VA

1d•Remote

About The Position

The Data Scientist designs and builds the v1 rule-weighted composite scoring logic that turns normalized risk signals into a transparent, defensible score. This role also prepares the scoring approach and model architecture for future interpretable ML-based scoring—ensuring explainability is preserved for adjudicator-facing workflows and audit needs. The ideal candidate blends practical applied data science with strong judgment around interpretability, traceability, and operational usability.

Requirements

Must maintain an active TS/SCI security clearance
Bachelor's Degree and 8 to 10 years of experience; Master's Degree and 6 to 8 years of experience; PhD and 3 to 5 years of experience (in lieu of Bachelor’s degree, 6 additional years of relevant experience)
3–5 years of applied data science experience delivering scoring, ranking, or decision-support models.
Experience implementing interpretable approaches (rule-based systems, transparent composite scores, and/or explainability methods such as SHAP).
Strong Python skills, including scikit-learn and common data science workflows.
Hands-on experience with SQL for data analysis, feature development, and validation.
Ability to communicate scoring logic clearly to technical and non-technical stakeholders (including explaining tradeoffs between accuracy and interpretability).

Nice To Haves

Familiarity with adjudicative, compliance, fraud/risk, or other risk-scoring domains
Python (scikit-learn, SHAP)
Jupyter
SQL
Explainability-first mindset: prioritizes transparency, traceability, and defensibility.
Analytical rigor: validates assumptions, tests edge cases, and avoids “black-box” shortcuts.
Collaboration: works effectively with data engineers and application teams to ensure scoring is usable and production-ready.
Documentation discipline: produces clear, auditable artifacts that describe logic, drivers, and limitations.

Responsibilities

Build and tune v1 rule-weighted composite scoring logic using normalized inputs from the common risk-signal schema.
Define scoring framework components (feature groupings, weights, thresholds, guardrails, and handling of missing/partial data).
Create interpretable explanations for scores and drivers suitable for adjudicator review (reason codes, key contributing signals, and traceable logic).
Design the scoring architecture to support evolution from rules/weights to interpretable ML models while maintaining auditability.
Prototype and evaluate interpretable model classes and explanation methods (e.g., SHAP-based explanations, constrained/monotonic models where appropriate, and rule-based hybrids).
Partner with data engineering and application teams to productionize scoring logic (data inputs, contracts, output formats, and performance expectations).
Establish validation and monitoring approaches (basic model/score QA, drift indicators, and score distribution checks).
Document scoring methodology, assumptions, and limitations for stakeholder understanding and accreditation/compliance artifacts as needed.