05
M.S. in Computer Science at The George Washington University.
Hybrid deep learning model combining BiLSTM, Transformer, and CNN architectures to estimate the Remaining Useful Life (RUL) of aircraft engines using the NASA C-MAPSS dataset. The model incorporates CBAM attention to enhance focus on key sensor features, positional encoding to preserve the temporal structure of sensor sequences, residual dilated convolutions to capture long-range dependencies, and transformer blocks for contextual sequence modeling, achieving average RMSE of 18.88.
Weakly supervised Multiple Instance Learning (MIL) framework for crowd behavior anomaly detection in surveillance videos, leveraging I3D-based spatio-temporal features extracted from RGB and optical flow streams. Utilizes multi-stream neural architecture, incorporating attention, residual connections, dropout, and layer normalization, achieving an AUC of 0.9022 and outperforming baseline classifiers in precision-recall balance.
Data mining and analytics on NYC 311 service requests, involving large-scale preprocessing, feature engineering, and exploratory analysis across geography, time, and socioeconomic factors. Applies clustering, contrast pattern mining, and co-occurrence analysis to identify borough-level complaint signatures, localized hotspots, and cascading complaints. Includes an anomaly detection framework using rolling statistical baselines to identify event-driven complaint surges, and predictive models for complaint volume and resolution time.
Retrieval-Augmented Generation (RAG) system over PubMed records and the OpenAI API to generate disease predictions, and recommend medical tests and treatments based on patient symptoms, demographics, and medical history, incorporating agentic AI for context-aware reasoning and improved diagnostic accuracy.
SSD-Mobilenet V1-based traffic monitoring system to detect, count, and classify vehicles in real-time, addressing need for efficient traffic control. Implemented vehicle color and speed prediction, storing images of detected vehicles, and logging detection data to a CSV file for analysis.
Built and evaluated multiple machine learning models to predict customer churn using a telecom dataset. Applied feature engineering and preprocessing, and benchmarked classifiers including Logistic Regression, SVM, Decision Trees, Random Forest, Gradient Boosting, and ensemble methods.
