Data Scientist

Responsibilities:

• Work with business and product teams to define the decision, target, prediction grain, outcome window, leakage controls, and cost of errors.
• Use Python and SQL/PySpark to explore large datasets, engineer useful features, and identify missingness, imbalance, drift, and segment-level behavior.
• Develop strong baselines and fit-for-purpose ML/DL solutions across tabular data, NLP, documents, or images when they improve outcomes.
• Use time- and cohort-aware validation, relevant metrics, threshold optimization, error slicing, robustness checks, and explainability techniques such as SHAP.
• Package models for batch or API inference and integrate testing, Git, Docker, model registries, and CI/CD with engineering partners.
• Monitor model and business performance, drift, latency, and failures, and define retraining, rollback, and human-review paths.
• Build models for use cases such as underwriting risk, straight-through processing, claims triage, severity and fraud detection, renewal or lapse propensity, document intelligence, and service or demand forecasting.

Requirements:

• 1-5 years of hands-on experience in data science, machine learning, or applied AI, including at least one model moved beyond proof of concept.
• Strong skills in Python, SQL, pandas, NumPy, scikit-learn, statistics, feature engineering, imbalanced learning, and model evaluation.
• Experience with XGBoost, LightGBM, or CatBoost, plus PyTorch or TensorFlow for relevant deep learning, NLP, or computer vision use cases.
• Comfort with APIs or batch services, testing, Git, Docker, and cloud or MLOps fundamentals; experience with AWS, SageMaker, MLflow, or Spark is a plus.
• Ability to explain assumptions, trade-offs, and limitations clearly and collaborate effectively with domain experts, product managers, and engineers.

Nice To Have:

• Experience moving a model from problem framing and validation into a real pilot or production workflow.
• Strong understanding of baselines, leakage prevention, metric selection, threshold trade-offs, and segment-level error analysis.
• Examples of turning ambiguous business asks into measurable ML problems and influencing stakeholders with evidence.