Data Science

A multi-objective deep learning framework utilizing a Multi-Task Learning (MTL) network topology with a shared EfficientNet-B0 backbone to simultaneously resolve joint regression and classification heads for body fat percentage estimation.

An AI-powered computer vision web application built with Streamlit and PyTorch that automates waste sorting. Achieves high-precision classification of organic and recyclable waste using a custom Convolutional Neural Network.

An end-to-end production grade machine learning pipeline predicting probability of loan default using a soft-voting ensemble model. Optimized via F2-score cost analysis to catch 96% of actual defaulters while retaining a 94% approval rate.

An AI-powered insurance premium prediction engine utilizing age-based stacked ensembles to replace manual actuarial-table pricing with automated, sub-second data-driven premium estimates with full SHAP-based explainability.

A systematic, walk-forward validated quantitative equity strategy on the NSE Nifty 100 Universe. A cross-sectional PyTorch MLP ranking 96 stocks by expected 5-day market-relative outperformance, optimized via CVXPY.