Built a large-scale occupational coding pipeline (pandas/jieba/rapidfuzz + Sentence Transformer embeddings), improving precision & recall by 18% and enabling downstream clustering/active learning.

Labeled thousands of Reddit posts with DSM-based guidelines; wrote Python scripts for cleaning, inter-rater reliability checks, and baseline modeling to speed up fine-tuning & evaluation.

Architected a text-to-music retrieval pipeline using CLAP embeddings + ChromaDB + emotion-aware reranking; improved search efficiency by 80% and boosted top-10 relevant hit rate by 30%.

An interactive civic platform that makes Blue Sky LA climate projects visible, searchable, and engaging for the public.

A secure, anonymous online forum supporting 100+ survivors to share experiences and seek help.

A full-stack BI platform: upload datasets, generate visualizations and insights, and accelerate analysis with AIGC narratives.

PDF extraction + reporting automation, MSDS expiration alerts, and procurement forecasting to reduce manual effort by 40%.

Multi-objective optimization + feature importance to identify operational ratios and suppress side reactions, improving catalyst lifetime.

Time-series + feature engineering + CV/HPO for MDS-UPDRS progression forecasting; ranked Top 30% among 1,805 teams.