00:00Pickup

No emergency call lost to language, noise or overload

Rescuora sits beside the call-taker. It captions and translates the caller as they speak, flags the sounds behind the voice, suggests the next protocol question and drafts the CAD record. Everything runs on hardware inside your center.

Pre-launch · design-partner pilots for 2027 are open

Call 4127 · Portuguese00:26LiveExample

Transcript and translation

  1. 00:00Call-taker911, what is the address of your emergency?Read back in PT: 911, qual é o endereço da sua emergência?
  2. 00:03CallerPTMeu pai caiu e não está respirando.My father collapsed and he isn't breathing.
  3. 00:08CallerPTRua das Flores 128, apartamento 3.128 Rua das Flores, apartment 3.
  4. 00:14Call-takerIs he awake? Is he breathing normally?Read back in PT: Ele está acordado? Está respirando normalmente?
  5. 00:17CallerPTNão, ele não responde. Está fazendo um barulho estranho.No, he doesn't respond. He's making a strange noise.
  6. 00:22Call-takerI'll tell you how to do chest compressions. Lay him flat on his back.Read back in PT: Vou explicar como fazer compressões no peito. Deite-o de costas.

Sonara · sounds

  • CheckIrregular breathing sounds00:06

Protora · suggested next step

  • 00:10Read the address back to the caller and confirm it.
  • 00:19Your protocol card suggests starting compression instructions now.

CAD draft · call-taker confirms

Language
Portuguese, detected
Location
Rua das Flores 128, apt 3 · confirm
Incident
Unresponsive, abnormal breathing
Priority
Set by call-taker

ExampleScripted dialogue written for this page and played in your browser. It is not a real call, not real audio and not output from Rescuora's models. The call-taker makes every decision.

00:04What we are building toward

Design targets, stated plainly

Rescuora has no live centers yet. These are the targets the first release is built to meet, and pilots will measure them.

Design target

<1s

Caption and translation delay behind the caller's voice.

Design principle

0bytes

Of call audio sent outside your building by default.

Products

5

Listen, hear, guide, respond and drill, on one appliance.

Draft plan

2027

First design-partner centers run Rescuora in shadow mode.

00:09The problem

The call-taker is doing four jobs at once

What happens on the floor today
  • A caller speaks a language no one on shift speaks. An interpreter line takes time to join the call.
  • A panicked caller cannot say what is happening. The alarm, the glass and the gasping are in the audio, and nobody has time to listen for them.
  • The protocol card has the right next question, but the call-taker is also typing the CAD record.
  • During a surge, ten callers describe the same crash while the queue grows.
What Rescuora changes
  • Captions and translation appear under a second behind the voice, both ways.
  • Sound events and rising distress show as tagged alerts, always with a word.
  • The next protocol question is suggested; the call-taker asks it or skips it.
  • The CAD record drafts itself from what was said, and waits for confirmation.

00:22How a call flows

From pickup to dispatch, with the call-taker in charge

  1. 00:00

    The call arrives

    Audio is forked inside your network to the Rescuora appliance. Nothing leaves the building.

    Your phone system
  2. 00:01

    Captions start

    Lingora transcribes the caller and detects the language. If it is not English, translation starts on the next line.

    Lingora
  3. 00:03

    Sounds are flagged

    Sonara listens behind the voice for alarms, glass, gunshots and distress, and tags what it hears.

    Sonara
  4. 00:05

    The next question is suggested

    Protora matches what it has heard to your center's protocol and suggests the next question. The call-taker decides.

    Protora, call-taker
  5. 00:20

    The record is drafted

    Location, incident type and key facts fill the CAD draft. The call-taker confirms each field before it is sent.

    Protora, call-taker
  6. Dispatch

    Units go out

    Dispatch happens in your CAD, by your people, the way it does today. Rescuora never sends units.

    Your dispatchers

00:30Who it is for

Built for the people on the floor and the people who run it

On the floor

Call-takers and dispatchers

Less typing, fewer interpreter waits, and a second pair of ears on every call.

Running the center

Center directors

Language access and quality review without sending audio to a public cloud.

Running the systems

Public-safety IT

One appliance, standard interfaces to your phone system and CAD, and an audit log for every suggestion.

00:41Under the hood

Built on the NVIDIA speech and agent stack

Runtime on the appliance
Speech
Riva streaming speech recognition, translation and speech synthesis
Audio cleanup
Maxine audio effects
Sound events
NeMo-trained audio classifiers on TensorRT
Guidance agent
NeMo with Guardrails, served as NIM microservices
Serving
Triton Inference Server on L40S-class GPUs
Edge
Jetson Orin and IGX Orin with Holoscan
Why on-prem

Emergency audio is among the most sensitive data a government holds. Running every model inside the center keeps it under the center's control, keeps working when the internet link fails, and keeps latency inside a second.

Model updates arrive as signed packages. The center chooses when to install them.

00:52For integrators

Every suggestion is an event you can read

Rescuora publishes captions, sound events, suggestions and CAD drafts as a stream of JSON events on the appliance. Your CAD, quality-review tools and wallboards can subscribe to it.

Each event carries the call ID, the call clock and whether a person confirmed it. Nothing becomes a CAD field until the call-taker confirms it.

events.jsonl · from the call bus
{"call": "4127", "t": 3.2, "type": "caption", "lang": "pt",
  "text": "Meu pai caiu e não está respirando.",
  "en": "My father collapsed and he isn't breathing."}
{"call": "4127", "t": 6.0, "type": "sound", "label": "irregular_breathing", "status": "check"}
{"call": "4127", "t": 9.1, "type": "cad_field", "field": "location", "confirmed": false}

ExampleIllustrative event shapes. The API is in design and will change.

01:05Questions

Questions call-center leaders ask

A person on our product or engineering team answers every message.

Q01Does Rescuora talk to callers on its own?

No. The call-taker speaks to the caller. When a translation is read back in the caller's language, the call-taker triggers it and hears it.

Q02Does it replace interpreter services?

No. Rescuora gets the call moving in the first seconds. Your center can still bridge a human interpreter, and the transcript is ready for them when they join.

Q03Is any audio sent to a cloud?

Not by default. Speech, sound and guidance models run on the appliance in your building. Sharing reviewed data for model improvement is opt-in and governed by your agreement.

Q04Is Rescuora certified?

Not yet. Rescuora is pre-launch. It is designed to support your center's existing protocols and quality-assurance work; it is not a protocol publisher or a certifying body.

Dispatch

Bring Rescuora to one shift, in shadow mode

In a pilot, Rescuora listens alongside your call-takers without changing what they do, so you can judge it on your own calls.