Sepsis early-warning model
Gradient-boosted survival model predicting sepsis onset 6h earlier than the clinical baseline, deployed in a hospital pilot.
I'm Sam — a data science grad student working on interpretable machine learning for healthcare. I like models that explain themselves and plots that need no caption.
Gradient-boosted survival model predicting sepsis onset 6h earlier than the clinical baseline, deployed in a hospital pilot.
Live dashboard forecasting energy use across 40 university buildings; saved an estimated 8% on heating costs.
Fine-tuned compact language models for medical triage notes — 85% smaller, same accuracy.
Distilled clinical language model for triage classification, running on-prem for privacy.
Demand forecasting + optimization that cut "no bikes available" complaints by a third in simulation.
Building faithfulness metrics for post-hoc explanations in clinical models.
Federated learning pipelines across six hospitals without moving a single patient record.
Thesis: Bayesian hierarchical models for small-area disease mapping.
Okafor S., et al. — under review
Okafor S., Lindqvist A. — ML4H Workshop, NeurIPS
Okafor S., et al. — JAMIA Open
Okafor S. — undergraduate thesis, awarded department prize
Open to research collaborations, internships and interesting datasets. The weirder the data, the better.
sam@example.com