At BC Cancer, I built an end-to-end RAG pipeline in Python over research papers and improved recall@5 from 60% to 87% on a labeled evaluation set.
I also led an external validation of a deep learning risk prediction model, using SQL to build a clinical cohort and benchmark performance. I used subgroup analysis to identify data inconsistencies and manufacturer-related bias, then shared the findings in Power BI dashboards and documentation.
Previously, I built and maintained financial data pipelines at Yonyou Network Technology Co., Ltd and developed a mapping dictionary that improved migration efficiency by 40 percent. On a Computer Vision on Food 101 project, I evaluated deep learning models and robustness, including through Grad-CAM analysis.

