SI-FF-ArcNet: Identical Twin Face Recognition (IEEE Published)
Designed a hybrid deep learning architecture integrating Swin Transformer and InceptionV4 using multi-scale feature fusion. Built a 512-D ArcFace embedding pipeline with 5-fold cross-validation and advanced augmentation. Achieved 99.96% validation accuracy and 98.20% F1-score across 50 epochs. Implemented cosine similarity-based open-set recognition for biometric authentication. Technologies: Python, PyTorch, Swin Transformer, InceptionV4, ArcFace Loss, OpenCV.
RakthaVahini—Blood Donation Platform
Developed an AI-assisted blood donation and emergency response application with donor registration, emergency requests, Firebase backend integration, and UI/UX for donor search, notifications, and request management. Technologies: Android Studio, Firebase, Kotlin, Generative AI.
Movie Recommendation System
Built a content-based recommendation engine using TF-IDF vectorization and cosine similarity. Achieved 85% recommendation accuracy through similarity scoring. Technologies: Python, Pandas, Scikit-learn.
Real-Time Semaphore Detection System
Developed a real-time pose-based signal recognition system and deployed it on NVIDIA Jetson Nano for efficient edge inference. Technologies: Python, OpenCV, MediaPipe, Flask, Jetson Nano.