Reinforcement Learning @ SCOT

Mostly building out backend and ML services on Scipher - a SaaS tool for patient phenotyping, understanding physician behavior, and commercial targeting.• Led design and development of physician call planning module, saving 40 hours/week across team• Trained a decoder-only language model using PyTorch to predict patient-level medical history• Developed GPT-4-based SQL generation tool for natural language querying of our data, enabling non-technical users to independently and flexibly generate insights• Cut cloud DB compute costs 50% by optimizing data model for common query patterns• Created self-serve ML modeling pipeline to identify and predict the presence of rare diseases• Implemented medical cost analysis dashboard to evaluate key cost drivers for any indication

• Led analysis and ML model development for a wide variety of commercial targeting exercises and journal publications• Software development for ACTICS platform• Developed concept embedding pipeline for drugs and diagnoses, reducing exploratory analysis 50%

• Developed analyses for physician targeting exercises, patient journeys, and journal publications• Rewrote core computation code, reducing runtime 75% and saving >$50k annually on computeFormerly HVH Precision Analytics