Voice2Care Agent¶

Overview¶

Voice2Care is BrainSAIT's AI-powered voice agent for patient interaction automation. It handles appointment scheduling, patient triage, health queries, and routine communications in both Arabic and English.

Core Capabilities¶

1. Voice Interaction¶

Language Support: - Arabic (Saudi dialect) - English - Automatic language detection - Real-time translation

Voice Features: - Natural speech recognition - Text-to-speech synthesis - Speaker identification - Emotion detection

2. Patient Triage¶

Assessment Functions: - Symptom collection - Severity scoring - Urgency classification - Care pathway routing

3. Appointment Management¶

Scheduling Functions: - Availability checking - Appointment booking - Reminders - Rescheduling - Cancellations

4. Health Information¶

Query Handling: - Medication information - Preparation instructions - Directions and hours - General health questions

Architecture¶

graph TB
    A[Patient Call] --> B[Telephony]
    B --> C[ASR Engine]
    C --> D[NLU Processor]
    D --> E[Dialog Manager]
    E --> F[Response Generator]
    F --> G[TTS Engine]
    G --> B

    H[Healthcare Systems] --> E
    I[Knowledge Base] --> E

Use Cases¶

Appointment Scheduling¶

Dialog Flow:

Voice2Care: مرحباً، أهلاً بك في مركز الرعاية الصحية. كيف يمكنني مساعدتك؟
           (Hello, welcome to the healthcare center. How can I help you?)

Patient: أريد حجز موعد مع طبيب باطنية
         (I want to book an appointment with an internist)

Voice2Care: تمام، سأساعدك في حجز موعد. هل لديك طبيب مفضل؟
            (Okay, I'll help you book an appointment. Do you have a preferred doctor?)

Patient: لا، أي طبيب متاح
         (No, any available doctor)

Voice2Care: حسناً. المواعيد المتاحة: الأحد الساعة 10 صباحاً أو الاثنين الساعة 2 ظهراً. أيهما يناسبك؟
            (Okay. Available appointments: Sunday at 10 AM or Monday at 2 PM. Which suits you?)

Patient Triage¶

Triage Protocol:

1. Greeting & Identification
2. Chief Complaint
3. Symptom Assessment
4. Duration
5. Severity (1-10)
6. Associated symptoms
7. Red flags
8. History Review
9. Triage Decision

Urgency Levels:

Medication Reminders¶

Reminder Flow:

Voice2Care: مرحباً، هذا تذكير بموعد دوائك.
            حان وقت تناول الميتفورمين 500 ملغ.
            هل تناولت الدواء؟

            (Hello, this is a medication reminder.
            It's time to take Metformin 500mg.
            Have you taken the medication?)

Integration Points¶

EMR/HIS Integration¶

Functions: - Patient lookup - Schedule access - Encounter creation - Note documentation

Telephony Integration¶

Platforms: - SIP trunking - Cloud PBX - Contact center - WhatsApp Business

API Endpoints¶

Initiate Call:

POST /api/voice2care/call
{
  "patient_id": "123",
  "purpose": "appointment_reminder",
  "language": "ar",
  "scheduled_time": "2024-01-15T09:00:00Z"
}

Handle Inbound:

POST /api/voice2care/webhook
{
  "call_id": "call-456",
  "event": "incoming",
  "from": "+966501234567"
}

Dialog Management¶

Intent Recognition¶

Supported Intents: - book_appointment - cancel_appointment - check_results - medication_query - directions - symptom_report - billing_inquiry - speak_to_human

Entity Extraction¶

Common Entities: - date/time - doctor_name - specialty - symptom - medication - body_part

Conversation State¶

{
  "session_id": "sess-123",
  "patient_id": "pat-456",
  "intent": "book_appointment",
  "entities": {
    "specialty": "internal_medicine",
    "preferred_date": "2024-01-20"
  },
  "dialog_state": "collect_time",
  "context": {
    "last_visit": "2023-12-01",
    "regular_doctor": "Dr. Ahmed"
  }
}

Voice Technology¶

Speech Recognition (ASR)¶

Specifications: - Real-time streaming - Word error rate < 5% - Arabic dialect handling - Medical terminology - Noise robustness

Text-to-Speech (TTS)¶

Features: - Natural voice synthesis - Arabic pronunciation - SSML support - Speed/pitch control - Emotion expression

Voice Profiles¶

voices:
  arabic_female:
    name: "Sara"
    language: "ar-SA"
    gender: "female"
    style: "friendly"

  english_male:
    name: "Omar"
    language: "en-SA"
    gender: "male"
    style: "professional"

Performance Metrics¶

Escalation Handling¶

Human Handoff¶

Triggers: - Patient request - Low confidence - Complex query - Emotional distress - Emergency detection

Process: 1. Summarize conversation 2. Transfer context 3. Warm handoff to agent 4. Log escalation reason

Compliance & Security¶

Data Protection¶

• Call recording consent
• PDPL compliance
• Data encryption
• Access controls

Clinical Safety¶

• Triage protocols reviewed by physicians
• Emergency detection validation
• Regular protocol updates
• Outcome tracking

Configuration¶

Call Flow¶

call_flows:
  inbound:
    greeting: "ar_greeting_01"
    menu:
      1: appointment_booking
      2: results_inquiry
      3: general_information
      0: human_agent
    max_retries: 3
    timeout: 30

Triage Protocols¶

triage_protocols:
  chest_pain:
    red_flags:
      - radiating_to_arm
      - shortness_of_breath
      - diaphoresis
    action_if_red_flag: emergency
    default_urgency: urgent

Best Practices¶

Conversation Design¶

1. Clear, concise prompts
2. Confirm important information
3. Offer escape options
4. Graceful error handling

Clinical Safety¶

1. Conservative triage
2. Clear emergency protocols
3. Regular protocol review
4. Outcome monitoring

Patient Experience¶

1. Natural conversation flow
2. Cultural sensitivity
3. Language preference respect
4. Accessibility support

Related Documents¶

• Healthcare Overview
• ClaimLinc Agent
• DocsLinc Agent
• Digital Transformation

Last updated: January 2025