My Projects

This study extends VisionTrader by incorporating news information through two complementary paths: extracting FinBERT-based sentiment scores as additional input features, and fusing news embeddings with price-volume representations via Concatenation, Gating Mechanism, and Cross-Attention. Evaluated robustness of each fusion strategy across different time periods; ongoing research.

Addresses instability in DRL-based portfolio optimization by incorporating enhanced market features and technical indicators. Uses Vision Transformer (ViT) to capture cross-asset and temporal relationships, demonstrating superior performance and stability across different markets and time periods.

A cloud-native attendance management system built with Spring Boot backend and Terraform-managed GCP infrastructure. Features include JWT-secured APIs, automated CI/CD with Cloud Build, and deployment on Cloud Run with Cloud SQL integration.

This project uses FinLab's real-time data to design and backtest trading strategies. Results, including account performance and monthly returns, are compiled into HTML reports and delivered via LINE Notify.

Processes and analyzes PDFs and images, including tax forms, certificates, and bid documents.

A smart app integrated with a medicine box to ensure drug safety by checking interactions using real-time web scraping and cached databases. Features include medication reminders, intuitive UI, and app-controlled compartments to prevent misuse.