Employ Nearest Neighbor, Fuzzy Matching Algorithms, SAS and Python to perform Claim’s Data Linkage, validate the linked data and assist the center with required Statistical Analysis and Visualizations following HIPAA guidelines.

- Worked with the Financial Crimes department to improve the recall rate of a model that predicts the probability of a transaction being associated with Money Laundering activities using Python and advanced Machine Learning like Random Forest and SVM.- Collaborated with a team of interns to pitch the "MicroBranches" concept in a US Bank Innovation Department competition, achieving top 5 placement and presenting the idea to senior leaders at US Bank headquarters.

Developed a Probability to Default model for a prominent U.S. bank's loan portfolio, improving its performance by 6% over the previous model through Statistical Modeling and SAS. Also, Monitored Loss Forecasting Models to guarantee accurate predictions and communicated any issues to stakeholders.Further, streamlined operations, reduced inefficiencies, and improved overall workflow within the team using Python.

- Worked closely with stakeholders and data providers to utilize business expertise for feature engineering from data sources such as Experian Credit Bureau and in-house data (including loan structure, repayment schedule, address, and repayment history) and verified the validity of these data.- Utilized Machine Learning Algorithms to address the following business challenges:1. Income Estimation: Estimated Income using OVR Classification on data from Credit Bureau. This variable helped in mitigating Portfolio's risk by 18%.2. Credit Risk Model: Developed an underwriting model using Logistic Regression and Gradient Boosting Tree algorithm, resulting in a 5% AUC-ROC improvement over the previous model. Weekly monitored to ensure the quality of predictions.3. Customer Management Model: Constructed a model with PR-AUC of 73% for a dataset to identify bad customers with an imbalanced distribution of the target variable (<5%) using weighted Logistic Regression, Gradient Boosting, and Easy Ensemble methods, without resorting to interpolation or discarding any pertinent information from majority class.- Utilized computer science fundamentals to address the following tasks:1. Developed a Feature Store for two major data sources and automated its setup on Jenkins to streamline the workflow for multiple stakeholders.2. Enhanced the in-house Python modeling library to accelerate the model development process.3. Designed and implemented the automation of the Model Monitoring framework to eliminate the need for manual effort in generating model monitoring reports.

Image Classification & Object Detection: Mapping shop images to credit risk