Event-based Multi-modal LLM Portfolio Investment System

September 2025 – May 2026•Under CityU CS Research Mentoring Scheme

• •Built a robust data retrieval engine to collect and process multi-source financial options data
• •Co-developed "Option Query Language" (OQL) to facilitate precise, LLM-driven financial analysis
• •Developed an automated annotation pipeline for high-quality dataset creation in quantitative research

Explainability-Stable Unlearning / FYP (in progress)

FYP — in progress

• •Develop an Explainability-Stable Unlearning Algorithm that penalizes significant shifts in SHAP values for a hold-out "Stability Set" of customers
• •Simulate non-IID financial heterogeneity with Dirichlet partitioning to create a federated environment where clients have skewed distributions
• •Quantify the compliance-utility trade-off by measuring computational cost and accuracy loss required to maintain explainability stability