00:00Platform

One appliance, five channels, one audit log

Rescuora runs as a GPU appliance inside your center. Every product reads the same call bus and writes to the same audit log, so a caption, a sound alert and a CAD draft always refer to the same moment in the same call.

Appliance · center classExample
Runs
Lingora, Sonara, Protora
Seats
Sized per call-taker position
Network
Inside your ESInet or LAN
Updates
Signed, installed when you choose

ExamplePlanned configuration; sizes are set per center.

00:04Architecture

Six layers, from audio in to records out

Planned architecture for the first release
LayerWhat it doesBuilt withWhere it runs
Audio ingestForks call and radio audio from your phone systemSIP/RTP media fork, NENA i3 interfacesAppliance
SpeechStreaming captions, translation, read-back speech, speaker separationRiva, Maxine, TensorRTAppliance GPU
Sound and stateSound events and caller distress cuesNeMo-trained classifiers, TensorRTAppliance GPU
GuidanceNext-question suggestions, CAD draft, multi-call linkingNeMo, Guardrails, NIMAppliance GPU
Console and CADCall-taker view, CAD field push after confirmationWeb console, CAD connectorsWorkstations
Review and learningAudit log, QA review, opt-in feedback to modelsRAPIDS analytics, SimoraAppliance, optional central

00:09The flywheel

How each center makes the models better, without sharing raw calls

1 · Review

Reviewed calls

Your QA team marks where a caption, a sound tag or a suggestion was wrong. Corrections stay in your audit log.

2 · Synthesize

Synthetic cases

Simora turns patterns from opt-in corrections into synthetic training calls, so no real caller's voice is reused.

3 · Update

Signed updates

Retrained models come back as signed packages. Each center tests them in shadow mode before switching them on.

00:15Deployment

Three sizes, all on hardware you control

Edge
For
Small centers, vehicles, field posts
Hardware
Jetson Orin or IGX Orin
Runs
Lingora, Sonara, Fieldora
Planned
Center
For
One PSAP or EMS dispatch center
Hardware
Rack server with L40S-class GPUs
Runs
All center products
Planned
Region
For
Consolidated or statewide centers
Hardware
Clustered GPU servers, failover pair
Runs
All products, shared across sites
Planned

00:22Control plane

Your IT team decides what runs, and when

What the control plane shows
  • Model versions per product, with release notes
  • Languages and sound packs installed
  • GPU load per position at peak
  • Health of audio ingest and CAD connectors
  • Which suggestions were confirmed, skipped or corrected
center.yaml · excerpt
center: eastside-911
languages: [es, pt, fr, vi]
sound_packs: [core, fire]
updates:
  channel: stable
  window: Tue 03:00-05:00
  shadow_first: true
cloud_sharing: false  # opt-in only

ExampleIllustrative configuration; field names will change.

00:30Security

Designed for criminal-justice and health data from day one

Stays in the building

Audio, transcripts and records stay on the appliance. No vendor remote access without a time-limited ticket you approve.

Encrypted everywhere

Disk encryption at rest, TLS between every component, keys held by the center.

Every action logged

Who saw what, which suggestion was shown, and who confirmed it, kept in an append-only audit log.

Dispatch

See the platform on your own floor plan

We map your positions, phone system and CAD in one session and come back with an appliance size and a pilot plan.