Financial Risk & Fraud Intelligence (Mphasis)

At Mphasis I built fraud-detection and risk-scoring models for financial-services clients, replacing brittle rule-based checks with tree-based ensembles (XGBoost, LightGBM, Random Forest) over engineered feature pipelines. Alongside the risk models, I built predictive-analytics models and live streaming dashboards on investment data, so portfolio teams could act on signals without waiting on batch reporting.

This was my early-career ground. I owned the modeling end to end: feature engineering on messy financial data, model selection, evaluation, and handing usable signals back to the business. I learned to treat a model as a product the client has to trust, not a notebook, which is the same instinct I carry into forward-deployed work today.

1. Fraud-detection and risk-scoring models with tree-based ensembles and engineered feature pipelines, lifting precision ~15% over the legacy rule-based systems and cutting manual review volume.
2. Predictive-analytics models on investment data, surfaced through live streaming dashboards so portfolio teams acted on signals in near real time.

Rule-based systems miss the patterns they were never written for, and every false positive is an analyst's hour. Moving to learned models cut manual review while catching more of what mattered, and the streaming dashboards turned batch-delayed reporting into something teams could act on the same day.