My Projects
Real-Time Gaze Tracking System
Developed a device-agnostic, real-time gaze tracking solution for clinical assessments. My work improved tracking reliability by 25% and data quality standardization by 40% across diverse hardware.
Medical Imaging Analysis with CNN
Built a high-accuracy CNN model optimized with TorchScript and ONNX to speed up medical image diagnosis, achieving 94% validation accuracy and reducing inference latency by 40%.
Document Classification using CNN-RNN & OCR
Automated document classification with a hybrid CNN-RNN model and OpenCV. My solution increased operational throughput by 40% and achieved 90% classification accuracy.
Real-Time Video Fraud Detection
Devised a real-time anomaly detection system using an LSTM model to detect fraudulent behavior in ATM video feeds, reducing manual monitoring by 55% and improving detection accuracy to 90%.
Financial Time-Series Forecasting & NLP
Architected deep learning models for time-series forecasting, improving asset price prediction by 22%, and used fine-tuned BERT models for sentiment analysis, enhancing trading signal precision by 18%.
Anomaly Detection with Autoencoders
Applied unsupervised learning and autoencoder techniques to optimize ROC-AUC thresholds for transaction anomaly detection, reducing false positives by 30% and strengthening fraud detection.