Interpretable Block-Term Tensor Network (BTTN) for predicting future-onset (incident) atrial fibrillation from a single sinus-rhythm 12-lead ECG on MIMIC-IV-ECG: the glass-box (time x lead) factor parameters ARE the explanation, with a measured-faithfulness framework, at parity with a CNN. Patient-grouped CV, patient-bootstrap CIs.

Cyclist safety risk assessment application for Austin using Random Forest classification with SMOTE for class balancing. Interactive Streamlit web app predicts cycling incident severity based on time, location, and environmental conditions to promote safer cycling practices.

Production-ready traffic incident prediction platform for Bengaluru with 95.22% ML accuracy. 10 backend engines, 32+ APIs, real-time impact assessment, resource planning, and corridor intelligence. Docker-ready. Built with CatBoost, FastAPI, and 8,173 real incidents.

ML system to predict workplace safety incidents in manufacturing — analyzes sensor data, work patterns, and environmental factors to prevent accidents before they occur

🚀 SmartAlert AI - Production-ready adaptive ML system for intelligent incident prediction. Achieves 44% F1-Score on challenging datasets with 75% false alarm reduction. Features 118 sophisticated features and real-time adaptive learning.