Improvement of classification accuracy by 6.5% for Arsenic-prediction across the glacial aquifer using custom built non-Euclidean spatio-temporal modeling with Graph-neural networks in PyTorch after performing data-pipe lining using Apache Hadoop, & graph-visualization in Neo4J while working for High Performance Computing in Graph Neural Networks

19% reduction in RMSE compared to conventional SIR-models achieved by building salable time series surrogate model in TensorFlow, Keras using RNNs like GRUs & Bi-LSTMs, & transformer models like Temporal-fusion-transformer (TFT) working for the Bio-Complexity Institute at IU.

Portfolio management for clients in this AI-driven mutual funds investment platform leveraging financial modelling tools and methodologies

1. Coordinated and contributed in the AGILE SDLC life-cycle of over 3 supplier-side inventory management products right from data sourcing, data wrangling using MySQL and PostgreSQL, data-integration with cloud using AWS, containerization using Docker, version control using GIT.2. Development of ML (SVM, KNN, Boosted Random Forest etc.)application called MonitSole which helped label data-integration errors leading to reduction in errors by 150% within 7 months of operation