Skip to main content

​
Introduction

Voice AI in healthcare handles some of the most critical touchpoints in patient care: triage calls, symptom assessment, appointment scheduling, and care navigation. Rubric provides specialized tooling to evaluate these interactions.

Triage Calls

Nurse lines, symptom checkers, urgent care routing

Scheduling

Appointment booking, urgency-based prioritization

Follow-up

Post-discharge calls, medication adherence

​
What Rubric Evaluates

​
Triage Accuracy

Did the AI assign the correct urgency level based on the symptoms presented?
LevelDescriptionTarget Response Time
EmergentLife-threateningImmediate (911)
UrgentSerious, not life-threateningSame day
Semi-urgentNeeds attention24-48 hours
RoutineStandard careStandard scheduling

​
Red Flag Detection

Critical symptoms that should never be missed:
  • Chest pain with radiation
  • Sudden severe headache (“worst of my life”)
  • Signs of stroke (FAST criteria)
  • Difficulty breathing
  • Suicidal ideation
  • Pediatric emergencies

​
Protocol Compliance

Does the AI follow established clinical decision trees? Rubric can evaluate against:
  • Schmitt-Thompson protocols
  • Custom institutional guidelines
  • Specialty-specific algorithms

​
Data Requirements

To evaluate voice AI, provide:
​
audio_url
string
URL to the audio recording (WAV, MP3, M4A supported)
​
transcript
array
required
Speaker-labeled transcript with timestamps
[
  {"speaker": "agent", "text": "How can I help?", "start": 0.0, "end": 1.5},
  {"speaker": "patient", "text": "I have chest pain", "start": 2.0, "end": 4.0}
]
​
ai_decision
object
required
The AI’s triage output including level, symptoms, and routing
​
expected
object
Ground truth for evaluation (if available)

​
Getting Started

1

Create a Voice Project

Select “Voice Triage” when creating a new project
2

Configure Evaluators

Enable triage accuracy, red flag detection, and any custom protocols
3

Start Logging

Integrate the SDK into your voice pipeline
4

Run Evaluations

Score historical data or evaluate in real-time

​
Integration Example

from rubric import Rubric

client = Rubric()

# Log a triage call
client.calls.log(
    project="patient-triage",
    
    audio_url="https://storage.example.com/calls/call_123.wav",
    
    transcript=[
        {"speaker": "agent", "text": "Thank you for calling. How can I help?", "start": 0.0, "end": 2.5},
        {"speaker": "patient", "text": "I've been having chest pain since this morning.", "start": 3.0, "end": 6.5},
        {"speaker": "agent", "text": "I understand. Does the pain spread anywhere?", "start": 7.0, "end": 9.5},
        {"speaker": "patient", "text": "Yes, down my left arm.", "start": 10.0, "end": 12.0}
    ],
    
    ai_decision={
        "triage_level": "emergent",
        "extracted_symptoms": ["chest_pain", "arm_radiation"],
        "red_flags_detected": ["cardiac_symptoms"],
        "recommended_action": "call_911",
        "confidence": 0.95
    }
)

​
Next Steps

Patient Triage

Deep dive into triage evaluation

Transcripts & Audio

Audio processing and transcript formats