Nithin Maragoni

Data Scientist @TD

Toronto, ON, CA

Designed and deployed machine learning models in Python (scikit-learn, XGBoost, PyTorch) to predict customer churn, fraud risk, and product recommendation propensity, improving model accuracy by 18% over previous benchmarks.• Led end-to-end ML pipeline development on Databricks integrated with Azure Data Lake and Snowflake, enabling scalable training and inference on 100M+ transaction records.• Implemented feature engineering pipelines using PySpark and SQL to transform structured, semi-structured, and streaming data from Kafka and core banking systems.• Collaborated with business stakeholders to define key metrics and ML use cases, translating business objectives into data science solutions with measurable ROI.• Delivered a real-time fraud detection and scoring system in under 6 months, reducing false positives by 22% and enabling compliance with OSFI requirements ahead of schedule.• Automated model retraining and deployment via MLflow and Azure DevOps CI/CD, reducing manual intervention by 60% and ensuring models stayed current with evolving data trends.• Performed model explainability analysis using SHAP and LIME, delivering insights to compliance and audit teams to meet OSFI and internal governance requirements.• Designed and executed A/B tests for model-driven customer outreach campaigns, achieving a 12% lift in cross-sell conversion rates.• Integrated deployed models with TD’s enterprise APIs for real-time scoring in digital banking channels, reducing decision latency to under 200 ms.• Built Power BI dashboards for executives to monitor model performance, customer engagement, and portfolio risk metrics in near real-time.• Partnered with risk and compliance teams to ensure AI models adhered to fairness, transparency, and bias mitigation guidelines in the Canadian banking sector.• Delivered and productionized a portfolio of four ML models within 8 months, accelerating time-to-market for AI-driven customer insights and fraud prevention initiatives.