Developed SmartShield, a privacy-preserving Android application for real-time SMS fraud detection using Spiking Neural Networks (SNNs). Built an end-to-end edge AI pipeline integrating a snnTorch-based classifier, TF-IDF feature extraction, and PyTorch Mobile deployment, enabling all inference to run entirely on-device without cloud connectivity. Curated a hybrid dataset combining the UCI SMS Spam Collection with custom Indian fraud samples, including UPI scams, KYC fraud, OTP theft, phishing attacks, and delivery scams. The proposed SNN classifier achieved 99.15% accuracy, 98.54% precision, 98.18% recall, 98.36% F1-score, and 98.83% ROC-AUC on held-out test data. Comparative analysis against an ANN baseline demonstrated the neuromorphic efficiency of the approach, requiring ~444× fewer effective computations and achieving an estimated 14,205× reduction in energy consumption, making the system suitable for ultra-low-power mobile deployment. To improve robustness against evolving fraud patterns, the framework combines SNN predictions with rule-based keyword and phishing URL analysis. Developed the native Android application using Kotlin and Jetpack Compose with support for real-time message analysis and background SMS monitoring. Currently extending the system through ONNX Runtime Mobile migration to reduce model footprint and improve inference efficiency on resource-constrained devices. Developed as Team Lead during HackSecure'26, a national cybersecurity hackathon organized under ISEA Phase-III (MeitY, Government of India).

Team Project- Abhishek Kumar (SNN, Neuromorphic Computing), Hitesh Pratap Singh (Application development)