Hospital Intelligent Triage: Multi-Intent AI

Executive Summary

Hospital outpatient lobbies face two high-frequency question types daily: "Where is XX examination/service?" and "What department should I visit for my symptoms?" Traditional solutions involve multiple triage nurses working in shifts, but peak hours still see long queues, low efficiency, and inevitable nurse fatigue.

This article demonstrates how to build an Intelligent Triage Agent using Tencent Cloud ADP, leveraging multi-intent recognition to automatically identify patient needs and route to different workflows:

• Location Query: Directly output location guidance for examination rooms/service windows
• Medical Consultation: Collect age, gender, and symptoms → Recommend department → Check if appointment needed → Recommend on-duty doctors

Key Highlights:

• Multi-intent recognition covering multiple scenario types with one Agent
• Slot-filling mechanism ensuring complete information collection
• Knowledge base maintains weekly schedules for doctor recommendations

1. Scenario Analysis: Two Core Needs at Triage Desks

1.1 Demand Distribution

Analysis of one week's triage records at a hospital outpatient lobby shows patient inquiries mainly fall into two categories:

1.2 Traditional Pain Points

2. Agent Design: Multi-Intent Recognition + Routing

2.1 Overall Architecture

                        Patient Question
                              ↓
                    ┌─────────────────────┐
                    │ Multi-Intent        │
                    │ Recognition         │
                    └──────────┬──────────┘
                               ↓
            ┌──────────────────┴──────────────────┐
            ↓                                     ↓
    ┌───────────────┐                     ┌───────────────┐
    │ Location      │                     │ Medical       │
    │ Query Flow    │                     │ Consultation  │
    │ Direct Output │                     │ Collect Info  │
    └───────────────┘                     └───────┬───────┘
                                                  ↓
                                          ┌───────────────┐
                                          │ Recommend     │
                                          │ Department    │
                                          └───────┬───────┘
                                                  ↓
                                          ┌───────────────┐
                                          │ Need          │
                                          │ Appointment?  │
                                          └───────┬───────┘
                                                  ↓
                                          ┌───────────────┐
                                          │ Recommend     │
                                          │ On-Duty Doctor│
                                          └───────────────┘

2.2 Multi-Intent Recognition

The Agent first determines which category the patient's question belongs to:

Example Dialogues:

Patient: Which floor is the CT room on?
Agent: Based on your query, please go to B1 Floor, Imaging Diagnostic Center, CT Room.

Patient: I've been having headaches lately
Agent: To recommend the right department, please tell me:
1. What is your age?
2. What is your gender?

2.3 Medical Consultation Flow Details

When identified as "Medical Consultation" intent, the Agent enters the information collection flow:

Step 1: Collect Basic Information (Slot Filling)

Step 2: Recommend Department

Based on collected information and preset triage rules, recommend the best-matching department:

Patient: I'm 18, male, my eyes are itchy and I can't see clearly.
Agent: Based on your symptoms, please go to 5th Floor, ENT & Oral Center, Ophthalmology Department for treatment.

Step 3: Check Appointment Need

After recommending a department, the Agent proactively asks about appointment needs:

• If patient says yes → Enter doctor recommendation flow
• If patient says no → End conversation or ask about other questions

Step 4: Recommend On-Duty Doctors

Query the department's schedule for the day and recommend available doctors:

Agent: Based on your symptoms, today you can register with Dr. Zhang, Chief Physician, at 5th Floor Area A, Ophthalmology Clinic.

3. Building with Tencent Cloud ADP

3.1 Create Agent and Configure Role

Role Configuration (Example):

You are an intelligent triage assistant for a hospital, helping patients with two types of questions:
1. Location queries: Provide specific locations for examination rooms and service windows
2. Medical consultation: Collect patient information and recommend appropriate departments and doctors

Important rules:
- You only provide triage services, no medical diagnosis
- Must collect age, gender, and symptoms before recommending departments

3.2 Configure Knowledge Bases

Location Information Knowledge Base (Example):

Schedule Information Knowledge Base (Example):

3.4 Configure Workflow

Intent Recognition Node:

• Input: Patient question
• Output: Intent type (Location Query / Medical Consultation)

Location Query Node:

• Input: Query content
• Process: Search location knowledge base
• Output: Location guidance

Medical Consultation Node:

• Input: Patient question
• Process: Slot filling (age, gender, symptoms) → Triage rule matching → Department recommendation
• Output: Recommended department + appointment need check

Doctor Recommendation Node:

• Input: Department name
• Process: Search schedule knowledge base
• Output: Today's on-duty doctor recommendation

4. Dialogue Examples

4.1 Location Query Scenario

Patient: Where is the parking lot?

Agent: Based on your query, please go to B2 Floor, Patient Parking Lot.

4.2 Medical Consultation Scenario (Complete Flow)

Patient: I've been having stomach pain lately

Agent: Please provide a valid age.
       Please specify your gender (male/female).

Patient: 28

Agent: Please specify your gender (male/female).

Patient: Female

Agent: Based on your symptoms, today you should go to 2nd Floor, Internal Medicine Clinic Cluster, Gastroenterology Department for further treatment.

Patient: Which doctor can I register with?

Agent: Based on your symptoms, today you can register with Dr. Zhang, Chief Physician, at 2nd Floor Internal Medicine Clinic Cluster, Gastroenterology Department.

5. Compliance & Boundaries

5.1 Strict Red Lines

5.2 Emergency Handling

When urgent symptoms are detected, direct to emergency immediately:

Patient: I suddenly have high fever and joint pain

Agent: Please provide a valid age.
       Please specify your gender (male/female).

Patient: 19, male

Agent: Based on your symptoms, today you can register with Dr. Zhang, Chief Physician, at 1st Floor Emergency Department.

6. Results

6.1 Core Metrics

6.2 Business Value

7. Extension Directions

7.1 Feature Extensions

7.2 Multi-Modal Interaction

Summary

This article demonstrated how to build a hospital intelligent triage assistant using Tencent Cloud ADP. The core design principles are:

1. Multi-Intent Recognition: One Agent covering "Location Query", "Appointment Query", and "Medical Consultation" scenarios
2. Slot Filling: Ensure collection of age, gender, and symptoms for medical consultations
3. Routing: Location queries output directly; medical consultations follow complete flow
4. System Integration: Connect with scheduling system for real-time doctor recommendations
5. Compliance First: Strictly distinguish triage service from medical diagnosis

Start now and build your intelligent triage assistant with Tencent Cloud ADP!

→ Start Free Trial

Related Reading

• Smart Triage Assistant
• Low-Code Multi-Agent Practical Guide
• Build Customer Service AI Agent in 6 Steps