CASE.STUDIES

Clinical decision support system designed to detect Silent Myocardial Infarction (SMI) in diabetic patients via Late Fusion of EHR and ECG data.

An architectural implementation and analysis of AlphaFold 2, focusing on the Evoformer attention mechanism and Invariant Point Attention (IPA) for 3D structure inference.

A hybrid forensic pipeline fusing ResNet50 for spatial artifact detection and LSTM for temporal inconsistency modeling to detect Deepfake videos with 90.2% accuracy.

A specialized Convolutional Neural Network designed to classify retinal optical coherence tomography (OCT) images into 4 diagnostic categories, specifically engineering solutions for severe class imbalance.

A normalized (3NF) PostgreSQL database engine designed to ingest, store, and query high-velocity NHL game telemetry, utilizing complex CTEs and window functions for tactical analysis.

A deep learning model analyzing over 320,000 crime records to predict incident types based on spatiotemporal features, outperforming logistic regression baselines.

An end-to-end Automatic License Plate Recognition (ALPR) system integrating YOLOv11 for object detection and Tesseract for Optical Character Recognition, optimized for real-world lighting conditions.

A hybrid forecasting engine that improves crime prediction accuracy by fusing historical police data with real-time social media sentiment analysis (Twitter/Reddit) using VADER and Random Forests.

A predictive analytics engine for construction sites that anticipates safety incidents by correlating quantitative labor data (fatigue) with qualitative superintendent notes (sentiment).

A high-efficiency route optimization engine implemented in Java, utilizing Min-Heaps and Adjacency Lists to minimize travel costs across complex weighted graphs.

A full-stack predictive system identifying at-risk telecom customers (95% accuracy). Containerized with Docker and deployed via a Streamlit dashboard for real-time business decision-making.