We've all been there: staring at a clunky, 10-year-old hospital web portal, clicking through endless nested menus just to book a simple check-up or download a PDF lab result. It's tedious, error-prone, and frankly, a waste of human potential. But what if you could just tell an AI, "Book me a dermatologist for next Tuesday and save my blood test results to my health folder," and it just... did it?
In this tutorial, we are diving deep into the world of autonomous agents, GPT-4o, and LLM-driven web navigation. By leveraging the revolutionary Browser-use library and Playwright, we’ll build a vision-capable agent that can navigate complex UIs, handle logins, and automate the most frustrating parts of healthcare administration. 🚀
Traditional automation tools like Selenium or Puppeteer rely on brittle DOM selectors (#button-id-342
). When a hospital updates its website, your script breaks. Using Browser-use with GPT-4o changes the game. Instead of looking for code, the agent sees the page like a human, understanding that a magnifying glass icon means "Search" regardless of the underlying HTML.
The system logic involves a feedback loop where the LLM perceives the browser state (screenshot + DOM tree), decides on an action, and executes it via Playwright.
graph TD
A[User Goal: Book Appointment/Download Report] --> B[LangChain Agent / Browser-use]
B --> C{Decision Engine: GPT-4o}
C --> D[Action: Click/Type/Scroll]
D --> E[Playwright Browser Instance]
E --> F[Hospital Portal UI]
F --> G[Visual & HTML Feedback]
G --> C
F --> H[Download Lab Report PDF]
H --> I[Structured Storage / RAG Pipeline]
I --> J[Task Completed ✅]
Before we start, ensure you have the following in your tech stack:
pip install browser-use playwright langchain-openai
playwright install
The core of our solution is the Agent
class from the browser-use
library. It wraps the browser interactions into a "thought-action" loop.
from browser_use import Agent
from langchain_openai import ChatOpenAI
import asyncio
async def run_healthcare_agent():
llm = ChatOpenAI(model="gpt-4o")
task = (
"1. Go to 'https://portal.city-hospital.com' and login. "
"2. Navigate to the 'My Appointments' section. "
"3. Find the first available slot for 'General Practitioner' next week. "
"4. Then, go to 'Lab Results', find the latest PDF, and download it."
)
agent = Agent(
task=task,
llm=llm,
)
history = await agent.run()
print(history[-1].result)
if __name__ == "__main__":
asyncio.run(run_healthcare_agent())
Healthcare portals often use complex authentication. The magic of Browser-use is that it can "read" the screen. If it encounters a captcha, it can notify the user or use vision-to-text to solve simple ones.
For handling files, we can extend the agent's controller to ensure downloads are routed to a specific directory for our RAG (Retrieval-Augmented Generation) system.
from browser_use import Agent, BrowserConfig
from browser_use.browser.context import BrowserContextConfig
config = BrowserConfig(
headless=False, # Set to True in production
disable_security=True,
extra_chromium_args=["--disable-web-security"]
)
context_config = BrowserContextConfig(
save_downloads_path="./medical_records_raw"
)
agent = Agent(
task="Navigate to the health portal and download the March 2024 Lab Report.",
llm=ChatOpenAI(model="gpt-4o"),
browser_config=config,
browser_context_config=context_config
)
Once the agent downloads the lab report, it’s just a "dumb" PDF. To make it useful, we process it into a vector database. This allows you to ask questions like, "Are my iron levels trending upwards compared to last year?"
While this tutorial focuses on the retrieval (the agent), the processing is where things get truly sophisticated.
Building a prototype is easy, but making a production-ready agent that handles edge cases—like session timeouts, dynamic pop-ups, and multi-factor authentication—requires a more robust architectural pattern.
For advanced implementation patterns on scaling these autonomous workflows and integrating them with secure healthcare data pipelines, I highly recommend checking out the technical deep-dives at ** WellAlly Tech Blog**. They cover great production-grade examples of how to wrap these agents in FastAPI and secure them for enterprise use.
We are moving away from an era where we adapt to software, and into an era where software adapts to us. By combining GPT-4o’s vision with Browser-use, we’ve effectively given our AI a pair of eyes and a mouse.
Next Steps:
What are you planning to automate next? The DMV? Your tax portal? Let me know in the comments below! 👇