Zymr developed a Python-based ML engine using scikit-learn, Pandas, and NumPy to train a binary classifier on historical EMR datasets. We exposed the model through a FastAPI microservice, enabling real-time scoring and integration into hospital dashboards. Feature importance was visualized using SHAP to ensure explainability for clinical teams. A retraining pipeline was implemented using Airflow to update the model weekly.
A U.S.-based healthtech SaaS provider serving hospitals and clinics with EMR-integrated platforms. The client aimed to leverage machine learning to predict patient readmission risk and optimize discharge planning.
Hospitals faced rising penalties due to high readmission rates under CMS guidelines. The client needed a predictive model capable of learning from EMR data, including clinical notes, vitals, and medication histories. The existing system lacked infrastructure for real-time prediction and model retraining.
The model achieved 87% prediction accuracy and enabled proactive interventions that reduced hospital readmissions by 28% within six months. Hospital partners reported improved care planning and better patient outcomes.
Zymr developed a Python-based ML engine using scikit-learn, Pandas, and NumPy to train a binary classifier on historical EMR datasets. We exposed the model through a FastAPI microservice, enabling real-time scoring and integration into hospital dashboards. Feature importance was visualized using SHAP to ensure explainability for clinical teams. A retraining pipeline was implemented using Airflow to update the model weekly.
Show More
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)