Ken Su Email & Phone Number

New York, New York, United States

Working on GenAI products at Dupont via Mondo

New York, New York, United States

Full-stack development of cloud-native data solutions for enterprise cloud transformation and operational efficiency.

New York, New York, United States

- Engineered an ETL pipeline for e-commerce merchant datasets (~100GB) in heterogeneous formats with PySpark and Spark SQL, enhancing data quality through advanced cross-joining techniques and Levenshtein distance matching.- Extracted product image embeddings with fine-tuned VGG16 model and developed an end-to-end sequential neural network pipeline with PyTorch and Keras to predict product sale volume at different timepoints by combining tabular product features and product image embeddings; the result R-square improved 87% from baseline XGBoost model in production.- Deployed the model as an end-to-end pipeline with virtual compute engines and buckets on Google Cloud Platform.

Hartford, Connecticut, United States

- Utilized Python's Seaborn and Matplotlib libraries to visualize by-variable density, action rate distributions, and feature correlations, optimizing variable encodings and boosting model's referral action capture rate from 32.4% to 43.1%.- Crafted and deployed an enhanced risk assessment model using LightGBM, augmented with rule-based triggers and fine-tuned via Optuna, achieving a 5.4% reduction in claim costs in backtesting, equating to an estimated annual savings of $15 million.- Streamlined data operations by automating an end-to-end model pipeline with Airflow, deploying on Amazon EC2 and S3 for enhanced scalability and efficiency, while ensuring robust version control and collaboration using Github.

Atlanta, Georgia, United States

- Devised and validated a novel Survival Analysis methodology to identify post-cryptogenic stroke patients at risk for atrial fibrillation using EHR data.- Architected an ETL pipeline using AWS Redshift to manage 1.8M+ EHR records, creating a pivotal dataset for atrial fibrillation research. This dataset now underpins numerous clinical studies, significantly advancing the field of effective stroke prevention. - Developed predictive ML modeling pipeline to set a new industry benchmark for accuracy in AF prediction. The work was accepted by International Stroke Conference 2023 for presentation.

Master Of Science - Ms, Data Science

Bachelor'S Degree, Computer Science & Economics

Activities and Societies: Phi Eta Sigma