The handover is where clinical data dies

Watch any emergency department entrance for an hour. Crews arrive, and the most critical information of the whole mission is transferred in thirty seconds of speech, from a tired responder to a busy nurse, over a stretcher, in noise. Then everyone moves on. Some of it gets written down. Some of it, later, from memory. Some of it never.

That thirty seconds is the most information-dense moment in emergency care, and the least recorded. Following conversations at HLTH Europe with the leadership of a large university hospital group, we spent the last weeks turning SONARA's core into a precise answer to that moment. This is what it looks like, stripped of any local specifics.

One pipeline, five verbs

SONARA runs as a single on-device pipeline during the mission: it captures, analyzes, computes, alerts and reports, in that order.

• Captures. The crew speaks or taps at the patient's side, hands-free, offline. Talking is the interface.
• Analyzes. The record is structured on the device into a clinical framework (ABCDE). Nothing is sent to a cloud.
• Computes. Vital signs, validated scores (NEWS2, GCS, qSOFA, shock index and more) and weight-based drug doses, at the point of care.
• Alerts. Deterioration flags raised early, with the relevant standing order surfaced as a reference prompt.
• Reports. A structured ATMIST or SBAR handover, ready when the doors open.

One principle above all of it: the clinician is the author and validates every record. SONARA does not decide. That single sentence settles most governance conversations before they start, and it is also why active, patient-specific decision support is deliberately a separate, later module with its own regulatory pathway.

From speech to the hospital record

A clean handover that stays on a tablet is only half the job. The record has to land where the hospital actually works: its EHR.

At handover, the validated record becomes a standards-based FHIR R4 bundle: a Patient, an Encounter, the vital signs and scores as LOINC-coded Observations, and the narrative as a DocumentReference. The clinician reviews it, then it posts to the hospital system as a single transaction. We have proven this write end to end on a major EHR vendor's sandbox: the note appears in the patient's chart, and nothing changes on the hospital side. No integration project, no middleware, no new screen for the ED.

That last property matters more than any feature. Every hospital we talk to carries scar tissue from integration projects. "Nothing changes on your side" is not a compromise; it is the design goal.

The patient who does not speak your language

A patient, a relative or a bystander often does not speak the crew's language. SONARA transcribes and structures the record in the language actually spoken during the intervention, across 40 locales, so a clean handover still reaches the receiving team when language would otherwise be a barrier. In a European capital, this is not an edge case either.

Configured to the service, not the other way round

Every deployment is configured with the service: their standing orders and local pathways, their formulary with doses aligned to local policy, their score set and thresholds, their languages, their handover routing. Every change is captured as an auditable trail. The product adapts to the service's governance; the service does not adapt to the product.

Why this order of operations

We could have started with dashboards, prediction, or triage AI. We started with the record, because everything else in the chain inherits its quality. Scores computed on clean data are trustworthy. Operational analytics built on structured missions are free. Interoperability with a record designed for standards is a write, not a project.

The handover is where clinical data dies today. It is also the single point where fixing the record fixes the chain.