Featured Posts
Projects in AI/ML, Data Analysis, Database Management, and Financial Modeling
🎬 GUI-ASFormer: Transformer-Based GUI Video Segmentation
This project adapts advanced transformer-based temporal models to analyze screen recordings and detect subtle human-computer interactions such as clicks, scrolls, and inputs within graphical user interfaces (GUI). Built upon the ASFormer architecture, the model is fine-tuned for GUI-specific semantics, enabling accurate segmentation and temporal localization of key user actions.
🌪️ Tornado Interactive Dashboard
An interactive data visualization platform that maps historical tornado activity across the U.S. using NOAA records. The dashboard empowers users to uncover spatiotemporal patterns in frequency, severity, and economic impact of tornadoes through dynamic filtering and rich, responsive charts powered by Altair.
📊 Interactive Visualization of School Demographics and Performance
This project provides an intuitive visual exploration of educational disparities by correlating school demographics, socioeconomic indicators, and academic performance across New York State. Through coordinated interactive views (maps, bar charts, gender pyramids), users can examine how race, income level, and school size affect student outcomes—offering insights for educators and policymakers.
📈 Robust Dependence Modeling and Copula Simulation
This project focuses on advanced statistical modeling using copula functions to capture non-linear dependencies between financial assets. By decoupling marginal behavior from joint distribution structures, the simulation framework enables more accurate stress testing and tail-risk analysis in portfolio risk management.
📀 Grace Hash Join Implementation
A low-level simulation of the Grace Hash Join algorithm designed for efficient execution of relational joins on large datasets under memory constraints. The project replicates real-world database behavior, including disk I/O, buffer pool usage, and recursive partitioning, offering hands-on insights into external join processing and system-level optimization.
🧬 Unsupervised Learning Core: K-Means, GMM, and CVAE Implementations
An in-depth exploration of unsupervised learning techniques implemented from scratch. It includes intuitive clustering via K-Means, probabilistic modeling with Gaussian Mixture Models (GMM), and representation learning using Conditional Variational Autoencoders (CVAE)—all visualized through interpretable 2D embeddings.
🖼️ CNN-RNN: Image Classification & Captioning from Scratch
A pure NumPy-based implementation of convolutional and recurrent neural networks for end-to-end image understanding. This project demonstrates fundamental deep learning concepts by classifying images using CNNs and generating descriptive captions with RNNs—without relying on high-level libraries like TensorFlow or PyTorch.