Feihan Sun

Built SQL and R pipelines to analyze 1M+ EHR records for studying links between stroke, hypertension, and Alzheimer’s disease (AD). Modeled drug exposure effects (Vorapaxar, NRTIs) on AD incidence using survival and regression models. Applying statistical programming (R, SQL, Python) and epidemiological methods to generate real-world evidence supporting research in neurodegenerative disease and pharmacoepidemiology.

Developed Python automation pipelines to extract and cluster survey data from the All of Us Program (~500k responses). Used text mining, semantic clustering, and encoding to convert free-text responses (alcohol use, smoking, activity) into quantitative features. Combined survey features with EHR-derived clinical variables to model behavioral risk factors of stroke recurrence.

Designed and implemented an autoencoder-based anomaly detection model, successfully distinguishing between normal and abnormal sound events. Used to streamline installation process of automobile equipment and ensure quality assurance. Converted raw .wav files into spectrograms (STFT) and applied data augmentation and hyperparameter tuning, reducing false detection rate by 15%. Delivered a robust anomaly detection pipeline that streamlined the installation QA process.

Co-developed Stage-Aware Learning (SAL), a dynamic treatment optimization method for COVID-19 clinical decision-making. Translated and validated Python algorithms into R, making the methods more accessible and reproducible for statistical researchers, thereby broadening the impact of the work. Benchmarked SAL vs. competing methods (BOWL, AIPW-Classifier), showing >10% accuracy improvement in simulation and real datasets. Collaborated across statistics and medicine to make results interpretable for clinical researchers.

Applied Six Sigma DMAIC to reduce WLCSP packaging defect rate by 8%, increasing yield and manufacturing efficiency. Conducted root-cause analysis and implemented process improvements, boosting product reliability and customer satisfaction.

Extended Nim game analysis from 2D to 3D/higher dimensions; implemented visualization tools for combinatorial game theory. Built reusable Python frameworks for algorithmic exploration of multi-dimensional game outcomes.