Home/ Docs/AI interpretation

AI interpretation

How rule-based flags and AI (LLM) interpretation work, and which plan unlocks each

4 min read

Two layers of interpretation, in two different places.

Rule-based flags — on the gateway, every ECG, every plan

Deterministic flags run on the gateway at ingest time, before the record is queued for cloud sync. Every ECG gets them; no plan check, no internet round-trip, no extra cost. Computes:

  • Heart rate band (bradycardia < 60, tachycardia > 100)
  • PR interval (1° AV block > 200 ms)
  • QT / QTc (long QT > 460 ms)
  • ST elevation / depression (lead-by-lead threshold)
  • Axis deviation (left / right / extreme)

Each flag carries a severity (info / borderline / alert). Flags are written into the HL7 OBX segments delivered to your HMIS, into the FHIR Observation.interpretation field, and onto the ECG record stored in the cloud.

The cloud also has a rule engine — but it's a backstop. It only fires if a record arrives without findings (e.g. legacy uploads or older gateway versions). For modern gateways, the cloud just reads what the gateway already computed.

AI = LLM augmentation — in the cloud, plan-gated

When this doc and the UI say "AI", we mean the LLM. It's the only AI surface in the product. Triggered when:

  • A clinician clicks Run AI analysis on the ECG detail page, or
  • AI_AUTO_RUN=true in the cloud-managed gateway config (auto-runs on every record).

The LLM receives the rule flags, measurements, and a 12-lead waveform summary, and returns a one-paragraph clinical narrative in plain English. Latency: 3–8 s.

Plan availability

Plan Rule flags (gateway) AI / LLM (cloud)
Basic
Professional
Enterprise
Multi-site

Rule flags are baked into the gateway, so every tenant gets them regardless of plan. AI / LLM access is gated by the ai_interpretation flag on the plan — Basic doesn't include it, every paid plan does.

What the AI is NOT

  • Not a diagnosis. Output is labelled "AI assist — clinician review required" in the UI and HL7.
  • Not a regulatory device. We do not claim FDA/CE clearance for the LLM output.
  • Not a replacement for the rule flags — they always run too.

Cost

LLM calls are metered: each tenant's monthly call count is visible at /admin/billing/usage (OFFIN) and /usage (facility admin). Plans include a generous monthly bundle; overage is billed per 1 000 calls.

Disabling

Set AI_AUTO_RUN=false in cloud-managed gateway config to turn off auto-runs. The manual Run AI analysis button is hidden on Basic.

Roadmap

Cloud-side cohort rules — comparing a record to the same patient's prior ECGs, or to a tenant-trained baseline — are on the roadmap but not yet implemented. They'd run alongside the gateway rule engine, not replace it.