ePlus Technology, IncArea Practice Director- AI & Machine Learning
May. 2019 - Sep. 2020Hartford, Connecticut Area- Focused on surfacing actionable insights from Electronic Health Records (EHR)
- Designed/Architected/Programmed a readmission classifier 'Appliance'- a complete workflow (written in python and using Apache Airflow for orchestration) that leverages ML and Deep Learning algorithms to predict the likelihood of a 30-day readmission
- Appliance is a completely air-gapped solution that ships preinstalled on moderately priced GPU accelerated OEM hardware (Cisco c240, NetApp HCI)
- Target market includes community hospitals and urgent care centers that don't have millions of dollars to invest in the necessary engineering and tech resources required to build a comparable solution (but do have an IT budget)
- Managed and grew a team from 1 Data Scientist, to 7, including software engineers, hardware engineers, developers and testers
- Appliance is completely ‘EMR Agnostic’ and can work with any EMR (Epic, Cerner, Allscripts, etc) to amplify the value of the data stored in the EMR Tech Highlights include:
- Intensive feature engineering from unstructured notes filed and stored in the EMR
- Features are then combined with other data in the EMR and fed into an Automated Machine Learning pipeline
- Output (probability of a 30-day readmission, on a per/patient-level) is then appended to the patient record and written back to the EMR
- Algorithms leverage Shallow Neural Networks (word vectors) and Latent Dirachlet Analysis ('LDA') to identify thematic information related to Readmissions
- Includes Interactive Visualizations (plotly+Dash/Flask) to provide non-technical users with the ability to investigate themes associated with patient readmissions
- Authored marketing collateral, proposals and social media messaging to help evangelize the Appliance