Eric Braun

• Founding individualized intervention targeting with causal AI/ML effort for improving utilization management and clinical quality
• Introducing deep learning driven forecasting to improve non-performant naive financial forecasts
• Building CI/CD AI/ML pipeline for a large language model application to be used as a template for AI/ML CI/CD.

End-to-end AI/ML architect and data science practice leader for improving Kaiser Permanente pharmacy workflows and clinical outcomes.

• Leveraged A/B testing and causal machine learning to build personalized multi-modal outreach targeting systems that optimized national CMS Five-Star medication adherence and mail order utilization outcomes.
• Developed foundational forecasting suites (Pytorch, gradient boosting and statistical methods) for Rx cost and volume, including over 20,000 different Rx's at multiple levels of geographic aggregation for strategic pharmacy staffing, inventory optimization and budgeting.
• Built LLM solution (Mixtral finetune) for creating structured data from free-text Rx sigs, facilitating
  patient safety and medication reconciliation workflows
• Architected KP Pharmacy's first AI/ML stack (MLOps, CI/CD and data pipelines) within on-prem requirements.
• On-boarded, managed and mentored three data scientists, leading them to deploy multiple
  production AI/ML models.
• Won 2023 KP National Data Science Competition: Predicting anti-psychotic medication adherence

AI/ML SME and end-to-end developer for clinical and operational workflows and strategic planning.

• Developed Pytorch DeepAR forecasting model for predicting daily patient emergency room and urgent care volumes to assist with COVID-19 surge staffing.
• Built a regional cardiac services ensemble forecast driven scenario analysis tool which guided SCAL cardiac catheterization, electrophysiology and cardiovascular surgery infrastructure planning; won 'Most Innovative' project at 2020 KP ML Conference
• Implemented a glassboxed Catboost regressor for predicting physician EMR messaging burden to drive SCPMG practice efficiency improvement.
• Created LightGBM surgical case length prediction model and user frontend for at Baldwin Park Medical center to improve operating room scheduling accuracy.
• Built daily, year ahead inpatient pediatric census forecast with prediction interval-based decision metrics to identify periods of high and low risk for over and underutilization

AI/ML and business intelligence support for Southern California Kaiser Permanente's clinical quality analytics team.

• Lead statistician for JAMA OPEN publication involving the development of two LightGBM O/E risk adjustment models, which revealed addressable variance in admission and length of stay outcomes in KP SCAL NICUs.
• Developed parametric distribution-based LOS goal target-setting methodology that allowed for apples to-apples comparisons across heterogeneous metrics and observation units.
• Created custom control charts based on BCa confidence intervals for robust process improvement monitoring of highly skewed metrics.
• Built multiple Tableau dashboards using complex data pipelines from the electronic medical record and other data sources, including the Hospital Quality Composite which is used for tracking inpatient care quality across the SCAL region.
• Initiated team's first code review and git version control workflows.

Predictive and prescriptive AI/ML decision support for university administration

• Created competing risks random forest model for student retention and graduation to assist multiple departments in decision making and student program improvement
• Developed XGBoost model to predict student application yield to assist enrollment management
• Conducted bootstrapped propensity score analysis to assess the effect of high course loads on student performance to guide GCSU's 'Complete College Georgia' initiative
• Coded SQL queries to retrieve data from Oracle databases to fulfill data requests from university administrative and academic entities
• Supervised and assisted two undergraduate interns create ordinal logistic regression and proportional hazard model based studies of the Supplemental Instruction program over six months
• Wrote executive summaries of the modeling projects to allow non-technical administration members understand and make use of developed models
• Presented modeling work at the 2015 USG Institutional Research Summit and 2016 University System of Georgia Summit