Enhancing Extubation Failure Prediction with LLM-Derived Features from Clinical Notes
Forthcoming work where I contributed to the feature extraction pipeline for deriving LLM-based features from respiratory therapy clinical notes.
Software Engineer, ML Data Foundation @ Stripe
Aditya Khowal
Currently
- Building Shepherd, Stripe's point-in-time correct feature engineering platform for batch training and online serving models.
- Work heavily with Spark, Flink and airflow to enable the freshest computed data to get to ML models in milliseconds.
- Writing Scala, Java, and Python for distributed systems around Stripe's Chronon fork.
Publications
Hybrid and distributionally robust optimization approaches for hospital blood bank inventory management
Presentation: Decoding nanopore signals for DNA data storage using predictive machine learning
Previously
Stripe
Architected and engineered a low-latency distributed system in Java and Go for real-time PII redaction, processing 300TB of daily log streams for GDPR compliance.
Amazon
Reduced screen rendering time on track transition by 80% by architecting a dynamic content update system for Alexa Music.
University of Missouri
Optimized hospital blood ordering policy using machine learning.
Deep learning for DNA basecalling from Nanopore ionic current disruption.