Amit Rai Chowdhury

Developing a product for anomaly and fraud detection. Involves a mix of approaches like clustering, Isolation Forest, Graph analysis, statistical approaches and business rules Developed an automated revenue forecasting system. An ensemble of approaches used - statistical models (ARIMA, SARIMAX), ML models (LightGBM, XGBoost), Bidirectional LSTMs (Tensorflow) Designed a system to detect defects in HVAC units using autoencoders (Pytorch) Developed a system to forecast loss in cooling capacity using TFTs

Successfully delivered a project aimed at reducing different types of paint defects for a prominent car manufacturer. Used multiple ML models like XGBoost, LightGBM, Logistic Regression, SVM and Random Forest Worked ith highkly imbalanced data using over and under sampling methods like SMOTE and Tomek links Delivered a POC to streamline the delivery of shipments from a trucking company with optimization techniques using Pulp

Implemented a solution for address matching using sentence embeddings and similarity search. Significantly reduced the turnaround time from 10-15 days to a few hours, and a significant savings in software license costs Designed a Recommender System using collaborative filtering for sales upselling and cross-selling. Delivered market-to-order strategy for sales teams in multiple geographies. Provided Qlikview dashboards to allow sales teams 360 degree visibility to their target market

Responsible for customer segmentation for marketing campaigns. Streamlined the process to reduce the delivery time from several days to one day. Automated the control group selection through Integer Programming reducing the effort from hours to minutes Delivered recommendations on product placement and bundling using Market Basket analysis Created a Purchase Propensity model using Logistic Regression to predict customer responses to upcoming campaigns.

Led a 30-member team working on diverse technologies across multiple geographies. Successfully handled two large portfolios - Compliance Reporting having over 20 subsystems, and Regulatory Reporting in Capital Markets having around 15 subsystems