Job Responsibilities:
- Design, develop, and support production ML solutions for Trade Surveillance and/or Financial Crime use cases (e.g., alert generation, prioritization/triage, risk scoring), with focus on measurable risk mitigation and control effectiveness.
- Apply supervised and unsupervised/semi-supervised methods (classification, anomaly detection, clustering; basic weak-supervision/heuristics where needed) to improve true-positive rates and reduce false positives.
- Execute the model lifecycle under guidance: problem framing, data sourcing and quality checks, feature engineering (behavioral/temporal and basic entity-relationship/graph features), model training, validation, calibration/thresholding, bias/fairness checks, monitoring, and refresh/retraining support.
- Contribute to model governance and risk management deliverables: documentation, test results, back testing, stability/drift analysis, and support for reviews with Model Risk / Audit / Controls partners (as applicable).
- Partner with Technology/MLOps to support CI/CD processes, model versioning/registries, and automated monitoring (data drift, performance) for reliable operation in production.
- Work with FCC / Surveillance SMEs, investigators/reviewers, and Operations to translate typologies/red flags into defensible ML controls; incorporate human-in-the-loop feedback to improve model usability and precision.
- Deliver interpretable outputs for end users: reason codes and explainability (e.g., SHAP/LIME-style drivers; simple counterfactual insights where appropriate) to support consistent alert dispositioning.
- Develop solution using GenAI/LLMs (e.g., summarizing narratives, extracting signals from unstructured text) as a complement to core statistical/graph ML detection methods.
Required Qualifications, Capabilities, and Skills:
- Master’s in a quantitative discipline (Computer Science, Statistics, Mathematics, Economics, Operations Research, or related).
- Minimum 3 years of hands-on applied ML / data science experience; exposure to Financial Crime (AML/sanctions/fraud) and/or Trade Surveillance is preferred.
- Strong Python and ML tooling (e.g., pandas, scikit-learn; Spark/PySpark a plus).
- Working knowledge of imbalanced learning and operational evaluation (precision/recall, PR-AUC, alert yield) and threshold optimization/calibration.
- Experience supporting model governance expectations: clear documentation, validation testing, benchmarking/baselines, back testing concepts, drift/stability monitoring, and explainability suitable for review.
- Strong communication skills to explain models and trade-offs, produce clear reason codes, and collaborate effectively with Compliance/Surveillance, Ops, and Technology stakeholders.