Curriculum Vitae

Designed and implemented a Dual-Tower Transformer (DT-Transformer) for in-hospital stroke mortality prediction using the multicenter eICU Collaborative Research Database. The architecture decouples categorical demographics from numerical vitals into separate encoding towers, achieving an AUPRC of 0.6171 (± 0.0058 over five seeds) — a 14.4% relative improvement over the strongest neural baseline. An Adaptive Runtime Safeguard clamps physiological outliers online within the data loader, ensuring stable, crash-free inference.

• Outperformed the strongest neural baseline, NN (AUPRC 0.5394), and a standard Transformer (0.5279); tree ensembles still lead (XGBoost 0.6467, Random Forest 0.6236)
• Implemented attention map visualization for clinical interpretability of feature importance
• Paper accepted and presented on May 24, 2026 at the 2026 6th International Conference on Computer Communication and Artificial Intelligence (CCAI)

• Collaborated closely with Head Advisor Dr. Kwong-Cheong Wong on deep learning for stroke mortality prediction using the eICU database
• Represented the association at CCAI 2026 in Nanjing, where the resulting paper received the Best Industrial Paper Award and the Best Presentation Award

• Processed 60 GB of multi-province power equipment inspection data to identify equipment fault frequencies and optimize inspection robot algorithms
• Contributed to the development of closing and splitting functions for power inspection robots to support real-world operational accuracy

• Reviewed four submissions and was subsequently invited to serve as a Program Committee member

• Engineered an end-to-end ML pipeline from raw eICU CSV ingestion (200,859 ICU stays) to model serialization, with reproducible preprocessing via custom DataLoader classes
• Modular architecture separating data, model, training, and evaluation layers for maintainability
• Multi-seed evaluation harness (5 seeds on a fixed patient-stratified split) with automated metric logging (AUROC, AUPRC, F1) across all baselines
• Released open-source at Dual-Tower-Transformer-eICU-Stroke and Data-Preprocessing-for-eICU

Multi-timescale LSTM for daily, weekly and quarterly power load forecasting, with an interactive decision-support system: 1.42% MAPE on daily predictions and 82% higher quarterly accuracy than ARIMA.

Full-stack web platform with a Vue.js front end, a Spring Boot REST back end and a Python microservice serving a trained scikit-learn classifier, with JWT role-based access control and Docker Compose deployment.

• Custom responsive static site from scratch — zero third-party UI frameworks
• Component-based layout via Jekyll includes, Liquid templating, and CSS custom properties (design tokens)
• Automated deployment via GitHub Pages CI/CD; SEO metadata, structured data, sitemap, robots.txt