Curriculum Vitae

Data Structures & Algorithms · Operating Systems · Computer Networks · Database Systems · Software Architecture & Design Patterns · Machine Learning & AI Fundamentals · Full-Stack Web Development · Cloud & Distributed Systems

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 (std 0.006) — a 14.4% relative improvement over the strongest neural baseline. An Adaptive Runtime Safeguard was integrated for out-of-distribution detection at inference time.

• Outperformed strongest neural baseline NN (AUPRC 0.5394) and Standard Transformer (0.5279); competitive with XGBoost (0.6467)
• Implemented attention map visualization for clinical interpretability of feature importance
• Paper accepted and presented at CCAI 2026 (The 6th International Conference on Computer Communication and Artificial Intelligence, May 24, 2026)

• Developed a computer vision pipeline for power inspection robots to automate circuit breaker state recognition, replacing manual visual inspection on high-voltage lines
• Fine-tuned YOLOv8 on domain-specific power equipment imagery, achieving reliable detection across varying lighting and weather conditions
• Built and maintained a 60 GB image dataset across Guangdong Province substations; standardized annotation formats and automated ingestion scripts
• Conducted fault-type EDA (Pandas / Matplotlib) to guide class-balanced sampling strategies
• Outcome: Written commendation from supervising engineer; earned 100/100 on formal internship evaluation

• Engineered an end-to-end ML pipeline from raw eICU CSV ingestion to model serialization, handling 200k+ patient records with reproducible preprocessing via custom DataLoader classes
• Modular architecture separating data, model, training, and evaluation layers for maintainability
• Stratified 5-fold cross-validation harness with automated metric logging (AUROC, AUPRC, F1) across all baselines
• Containerized full training environment with Docker; open-source release in preparation

• Designed and trained LSTM, GRU, and vanilla RNN models; best LSTM configuration achieved 1.42% MAPE on the held-out test set
• Built full preprocessing pipeline: missing-value imputation, normalization, sliding-window construction, leakage-free splits
• Benchmarked against MLP, Linear Regression, and XGBoost baselines with publication-quality visualizations (Matplotlib / Seaborn)
• Implemented early stopping, learning-rate scheduling, and gradient clipping to stabilize RNN training

• Built a full-stack system with Vue.js SPA frontend, Java Spring Boot RESTful backend, and embedded Python ML microservice via REST
• Integrated trained Scikit-learn classification model for on-demand inference without re-training overhead
• JWT-based auth with role-based access control; deployed via Docker Compose on Linux server

• 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