Research

MetaFlow, a metadata-driven AutoML framework that selects models using dataset properties combined with heuristic and constraint rules to eliminate unpromising algorithms before training.

An Automatic Facial Image Labeling System built upon semi-supervised learning methods using pre-trained models to accurately label facial attributes like ethnicity, emotion, age, and gender.

A real-time surveillance framework for automated violence detection using deep learning architectures (CNN-LSTM & Attention) tuned for fast and reliable scene-level threat recognition.