How Salus's AI personas actually work

"AI-native" does not mean AI making clinical calls. Inside Salus, intelligence is embodied as a team of seven named personas, each responsible for one part of the lab, each with a least-privilege permission scope, and each supervised by a human. AI assists people; it never replaces scientists or makes final clinical determinations.

Personas are real workers with defined jobs, not a chatbot layered on top of the software. They draft, flag, extract, and monitor. They never release a result on their own.

Mary handles accessioning intake, extracting manifests, label sheets, and CSVs into a Suspense Queue and pre-filling accessioning for a human to confirm — she is the one persona that runs autonomously within her scope, because intake pre-fill is low-risk and always reviewed before it becomes a real record.

Allyson flags anomalous and out-of-range results and drafts verifications for sign-off. Conrad monitors QA/QC, Westgard rules, and out-of-spec events, and keeps CAP/CLIA documentation current. Andre validates HL7/FHIR messages and LOINC/SNOMED coding before anything reaches a state feed. Judy tracks turnaround time, bottlenecks, and staffing, and drafts reagent reorders. Kristin watches for outbreak signals and flags cases for epidemiologist investigation.

Cher is the supervisor: she can pause, throttle, or down-scope any of the other six, but per design she can never fully turn one off, that decision is reserved for a human.

Every persona sits at one of four levels: Suggest, Acts with approval, Autonomous in scope, or Supervisor. Most personas operate as "Acts with approval," meaning they draft the work and a person approves it in one click, nothing is auto-committed.

A floating assistant orb appears on every screen and surfaces whichever persona owns that screen, with a proactive tip, a live operational pulse, and a one-click action, so the right worker is always one click away without anyone hunting for a menu.

Disabling a persona is a human-only action; the kill switch always stays in a person's hands. Personas hold scoped permissions that mirror staff roles, so a persona structurally cannot do something its human counterpart could not do. Every status, autonomy, and scope change is written to an append-only control log recording who changed what, and when, the same audit discipline applied to clinical actions applies to the AI layer itself.