Building AI for clinical settings is harder than it looks. It takes more than picking a good model or tuning the right parameters. When engineers enter a hospital ecosystem, they quickly discover that