Lattice Health

Hospitals now run clinical AI from multiple vendors for tasks like imaging triage, sepsis prediction, and documentation, often embedded inside PACS and EHR systems with no shared view of accuracy, drift, or subgroup performance. Vendors can silently swap the model behind an API overnight, shifting performance without the hospital's knowledge until a clinician notices. When compliance, counsel, or a regulator later asks what a model was doing on a specific date, assembling an answer can take weeks, if it is possible at all. There is no consolidated oversight layer even though these models are already making decisions that affect patients.

Lattice sits beside a hospital's existing AI stack, reading HL7 and DICOM signals the models already emit rather than requiring a rip-and-replace integration. It continuously measures drift (PSI), subgroup fairness disparity, latency, and detects silent vendor model swaps via statistical tests on confidence distributions. Each morning it delivers a cryptographically signed, per-role PDF report (for IT directors, CMIO/CIO, and compliance officers) rather than a dashboard, so leadership can read it in their inbox and verify it offline with standard tooling against the institution's own public key. It also rolls up one-click radiologist thumbs-up/thumbs-down feedback into model-level findings, and produces chain-of-custody evidence packets for later regulatory or legal scrutiny, aligned to IHE, HIPAA-safe egress, HHS §1557, and FDA PCCP standards.