Paul Yoon

SageMaker HyperPod Team

Joining the SageMaker HyperPod team for summer 2026: distributed training infrastructure for large-scale ML workloads.

Developing a framework for agents to execute autonomous machine learning research on the UK Biobank's massive dataset.

PI: Bao Do, MD

Built an automated radiology retrieval pipeline orchestrating 5+ external APIs (Gemini, Google Images, internal libraries), cutting ~90% of manual lookup time for clinicians. Wrote structured-text extraction to convert radiology prose into typed clinical features for downstream reasoning.

Shipped a Python forecasting component (Facebook Prophet) that outperformed the production model by 120% MAPE and was adopted directly into the analytics pipeline. Built a Spark SQL anomaly detector on S3-backed event logs to reduce false positives for power users.