自动化你的健康：构建具备 GPT-4 函数调用和 Google Calendar 的 AI 医疗代理

Source: Dev.to

请提供您希望翻译的完整文本（除代码块和 URL 之外），我将按照要求保持原有的 Markdown 格式并将内容翻译成简体中文。

介绍

我们都有过这种经历：身体不舒服，但想到要在笨拙的医院门户上操作，或者在电话线上等上 20 分钟才能预约医生，这种感觉就像又添了一个症状。

如果你只需要说，“我下背部有剧痛，帮我找一位下周二下午有空的骨科专家”，然后让智能 AI 代理处理剩下的事，会怎样？

在本教程中，我们将构建一个 Healthcare Agent Controller。通过利用 OpenAI Function Calling、AI agents 和 Google Calendar API，我们将创建一个系统，实现以下功能：

1. 理解自然语言请求。
2. 查询医生的可预约时间。
3. 与你的个人日程进行交叉对照。
4. 完成预约——全部通过一个简洁的 FastAPI 后端完成。

这就是医疗自动化和个性化 AI 助手的未来。

架构概览

要构建可靠的代理，我们需要在 “大脑”（LLM）和 “手”（API）之间进行明确的分离。下面是数据流图。

sequenceDiagram
    participant User
    participant FastAPI as API_Gateway
    participant OpenAI as GPT-4o_(Brain)
    participant GCal as Google_Calendar_API_(Hands)

    User->>FastAPI: "Book a dentist for next Monday morning"
    FastAPI->>OpenAI: Prompt + Tools Definition
    OpenAI->>OpenAI: Analyze Intent & Extract Date/Type
    OpenAI-->>FastAPI: Call Function: search_doctors(specialty, date)
    FastAPI->>GCal: GET /events (Check slots)
    GCal-->>FastAPI: Returns Available Slots
    FastAPI->>OpenAI: "Found 10:00 AM and 11:30 AM"
    OpenAI-->>FastAPI: Call Function: create_appointment(time)
    FastAPI->>GCal: POST /events (Book it!)
    GCal-->>FastAPI: Confirmation
    FastAPI-->>User: "Success! You're booked for 10:00 AM."

前置条件

注意: 函数调用允许 GPT‑4 描述您在代码中定义的函数的参数。我们将定义两个工具：一个用于获取可用时间，另一个用于创建事件。

1️⃣ 定义工具 (tools.py)

# tools.py
tools = [
    {
        "type": "function",
        "function": {
            "name": "get_calendar_availability",
            "description": "Check if a specific time slot is free on the doctor's calendar",
            "parameters": {
                "type": "object",
                "properties": {
                    "start_time": {
                        "type": "string",
                        "description": "ISO‑8601 formatted start time"
                    },
                    "end_time": {
                        "type": "string",
                        "description": "ISO‑8601 formatted end time"
                    },
                },
                "required": ["start_time", "end_time"],
            },
        },
    },
    {
        "type": "function",
        "function": {
            "name": "book_appointment",
            "description": "Create a new calendar event for the medical appointment",
            "parameters": {
                "type": "object",
                "properties": {
                    "summary": {
                        "type": "string",
                        "description": "Reason for the visit (e.g., 'Orthopedic consult')"
                    },
                    "start_time": {"type": "string"},
                    "end_time": {"type": "string"},
                },
                "required": ["summary", "start_time", "end_time"],
            },
        },
    },
]

2️⃣ Google Calendar 的辅助函数

为简洁起见，我们假设您已经完成了 OAuth2 流程，并且已有 service 对象准备好。

# gcal_helper.py
from googleapiclient.discovery import build

def create_appointment(summary: str, start_time: str, end_time: str):
    """
    Insert a new event into the authenticated user's primary calendar.
    """
    event = {
        "summary": summary,
        "start": {"dateTime": start_time, "timeZone": "UTC"},
        "end":   {"dateTime": end_time,   "timeZone": "UTC"},
    }
    # `service` should be a pre‑authenticated Google Calendar service instance.
    return service.events().insert(calendarId="primary", body=event).execute()

3️⃣ FastAPI 端点 (main.py)

# main.py
import os
import json
from fastapi import FastAPI
from openai import OpenAI
from tools import tools
from gcal_helper import create_appointment

app = FastAPI()
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

@app.post("/schedule")
async def handle_medical_request(user_input: str):
    """
    Receive a natural‑language request, let the LLM decide which tool to call,
    and execute the corresponding function.
    """
    # 1️⃣ Send request to OpenAI
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": user_input}],
        tools=tools,
        tool_choice="auto",
    )

    message = response.choices[0].message

    # 2️⃣ Did the model request a function call?
    if message.tool_calls:
        for tool_call in message.tool_calls:
            if tool_call.function.name == "book_appointment":
                args = json.loads(tool_call.function.arguments)
                result = create_appointment(
                    summary=args["summary"],
                    start_time=args["start_time"],
                    end_time=args["end_time"],
                )
                return {"status": "success", "data": result}

    # If no tool call, ask the user for clarification
    return {"status": "clarification_needed", "message": message.content}

4️⃣ Production Considerations

• HIPAA compliance – encrypt data at rest and in transit, restrict access, and audit logs.
• Retrieval‑Augmented Generation (RAG) – integrate a medical knowledge base for accurate advice.
• State management – handle multi‑turn conversations (e.g., confirming date/time).

For a deeper dive into production‑ready agent patterns, see the WellAlly Blog (specialized guides on prompt‑injection prevention, multi‑agent orchestration, etc.).

下一步

🎉 您已经构建了一个最小的、功能完整的医疗代理控制器！

通过将 GPT‑4 的推理能力与 OpenAI 函数调用的结构可靠性相结合，我们将繁琐的行政任务转变为流畅的对话。继续迭代，确保管道安全，您将拥有一个可投入生产的 AI 助手，只需几句话即可帮助患者预约。

使用医疗目录 API 或其他数据源来丰富专家查询功能。

您是在为医疗或生产力构建 AI 代理吗？在下方留言或分享您的仓库——我很想看到您是如何实现这些模式的！