The Savant: voice-first emergency guidance for untrained bystanders, powered by Vocal Bridge

A hands-free assistant that talks a bystander through CPR, hemorrhage control, and airway management, then files a structured incident report for the hospital, built on Vocal Bridge.

The problem: most people freeze when seconds matter

Picture a roadside accident in the middle of a desert. You are the only witness, the nearest help is far away, and the person in front of you is having a heart attack. You know that something must be done in the next few minutes, but you do not know what. Most people do not. The gap between "I want to help" and "I know the correct sequence of steps" is where lives are lost.

The Savant is built for exactly that moment. It assumes the user has no training and no time to read. It assumes their hands are occupied and their attention is on the patient, not a screen.

What Devam built

The Savant turns a phone into a calm, protocol-driven emergency coach. A panicked bystander opens a voice link and describes the situation, and instead of returning a wall of first-aid text, the agent asks one focused question at a time, adapts to each answer, and tells the person the single next correct action, escalating to emergency services when that is the right call.

That is the whole point. The value is not reference material, it is a conversation that meets a frightened bystander where they are and gives them one clear step at a time. The guidance is grounded in TCCC (Tactical Combat Casualty Care) and AHA (American Heart Association) protocols for procedures like CPR, hemorrhage control, and airway management. Voice fits because in an emergency your hands are busy and there is no time to read.

How The Savant uses Vocal Bridge

The Savant is built on Vocal Bridge's Voice for App surface. A cyberpunk-styled Streamlit dashboard shows the live connection status, a real-time conversation log, and system monitoring while the agent and the user talk. Vocal Bridge handles the real-time voice channel, the low-latency, sub-second back-and-forth, the streaming transcript, and the barge-in that lets the user interrupt the moment the situation changes.

The architectural split is clean. Devam owns the emergency-response logic, the protocol knowledge, and the post-call pipeline. Vocal Bridge owns the voice. After the call ends, The Savant chains Vocal Bridge AI processing with a Zapier MCP tool to transcribe the audio, restructure it into JSON with fields like severity and patient vitals, write a row to a Google Sheet, and send an incident-summary email to a supervisor.

from vocalbridge import Agent, on_call_end

agent = Agent(name="savant", protocols=["TCCC", "AHA"])

@on_call_end
def file_report(transcript):
    record = agent.summarize(transcript, schema={
        "severity": "string",
        "vitals": "object",
        "interventions": "array",
    })
    zapier_mcp.invoke("append_sheet_and_email", record)

Illustrative. The actual implementation structures richer incident fields and equipment details.

What's hard about voice here

An emergency call cannot stall, and it cannot keep following a script past the point where the situation changes. If the user reports a new symptom while the agent is still talking, it has to hear that and change course, not finish the line it started. Vocal Bridge's barge-in and streaming transcript are what let it pivot from one instruction to "call emergency services now" without missing a beat.

When the stakes are this high, that responsiveness is not polish. It is the whole product.