NEJM AI Grand Rounds Epic’s Approach to AI with Seth Hain
61 snips
Feb 18, 2026 
Seth Hain, Senior VP of R&D at Epic who leads work on foundation models and the Cosmos healthcare dataset. He discusses building scalable structured medical event models, how Cosmos balances data sharing and trust, and integrating generative AI into clinical workflows. The conversation covers localization of predictions, cautious deployment for interventions, and treating AI as infrastructure for safer care.
AI Snips
Chapters
Transcript
Episode notes
From 4‑H Computer Kid To Epic R&D Head
- Seth Hain traced his path from math and 4-H programming to joining Epic straight out of graduate school.
- He credits a great high-school teacher and early programming experience for steering him into healthcare software.
Longitudinal Records Power Real‑World Evidence
- Cosmos stitches de-identified records across participating health systems to form longitudinal patient journeys.
- That combined view enables real-world evidence and applications directly in clinical workflow.
Patient Stories As Structured Event Sequences
- Epic trained a transformer on sequences of structured medical events instead of natural language tokens.
- The largest trained model used ~1 billion parameters and 115+ billion discrete medical events for pretraining.