How to build a Patient Support Agent

Workflow Overview

This flow is designed to help a telehealth support assistant answer patient questions by combining information from official documents and electronic health records (EHR), then generating a helpful, policy-compliant response.

Step-by-Step Node Breakdown

1. Patient Message (Input Node)

• Purpose:
  This is where the patient’s question or message enters the workflow.

• Example Input:
  “Who is Marcus Lee’s primary physician?”

2. Official Documents (Knowledge Base Node)

• Purpose:
  Searches through official telehealth documents (e.g., policies, instructions) for relevant information to answer the patient’s question.

• How it works:

  • Receives the patient’s message as a search query.

  • Returns relevant document chunks (e.g., appointment policies, contact info).

3. EHR (Knowledge Base Node)

• Purpose:
  Searches through indexed EHR files for patient-specific or clinical information.

• How it works:

  • Also receives the patient’s message as a search query.

  • Returns relevant EHR data (e.g., physician assignments, visit history).

4. OpenAI (LLM Node)

• Purpose:
  Acts as the AI assistant that crafts a support response.

• How it works:

  • Receives the outputs from both the Official Documents and EHR nodes, as well as the original patient message.

  • Uses a carefully designed system prompt to ensure responses are:

    • Polite, professional, and empathetic.

    • Based only on provided documents and EHR (