At KU IoT R&D Lab × KyraWorks, I built and evaluated a class-agnostic retail item detection baseline on SKU110K. Across training iterations, I improved precision from 78.2% to 86.8% and ran robustness tests that identified texture noise as the largest failure mode.
On RupChitran, I built backend logic for facial recognition and emotion detection, curated the training dataset, and contributed to training the CNN-based emotion classifier. I also contributed to a BiLSTM attention model for football match prediction and built its Flask interface.
I built MockingBird, a bioacoustic forest-health monitoring tool that combines BirdNET detection with an ecological health index. I’ve also developed full-stack and desktop applications, including CodeMarga and Torrex, working across React frontends and Python-based backends.
