🎉 AI Engineers: Join our webinar on Prompt Engineering for AI Agents. Register here >>

April 7, 2024 - last updated
GenAI For Practitioners

The potential of LLMs in Web Agents: Going beyond basic scraping

Igal Leikin
Igal Leikin

Igal is a Senior Software Engineer at Aporia.

2 min read Nov 30, 2023

Setting the stage 

You’re familiar with LLMs for coding, right? But here’s something you might not have thought about – LLMs are your secret weapon for creating advanced Web Agents. Forget basic scraping. This post will explore where AI interacts with the web.

LLMs in web interaction

LLMs are becoming more than pretty decent at understanding code, thanks to their extensive training on diverse coding languages available across the internet. This makes them ideal for more than just GitHub Copilot or code reviews. Imagine using LLMs to in Web Agents that aren’t just an improvement but a complete transformation.

Let’s say you’re tackling sentiment analysis for a new product. The challenge? Reviews are scattered across 80 e-commerce sites. The old approach would be writing 80 parsers. But there’s a smarter way – a single Python script, all thanks to the combined power of Playwright and LLMs.

Beyond basic scraping with Playwright and LLMs

Here’s how it works. With Playwright and LLMs, you can manage all these websites using just one script. You ask the LLM to find a CSS selector for customer reviews in a piece of HTML. The LLM responds with the selector, and you’re set to collect reviews from a multitude of sites quickly and efficiently.

What we’re discussing here goes beyond standard web scraping. Imagine AI agents that can interact with web applications – clicking buttons, entering text, and navigating through pages. It’s like having an assistant who’s not just observing the web but actively engaging with it.

Endless possibilities of LLMs in web interactions

The applications? Limitless. Think about automatic UI testing that adapts and doesn’t break with every minor change. Or consider the ability to gather complex datasets from diverse web sources. This is where your creativity as an ML practitioner comes into play.

Sure, this sounds great, but it’s not without its hurdles. Ensuring accuracy across various web structures and making these agents adaptable is a tricky business. But hey, that’s what makes this field exciting, right?

As you ponder over your next project, consider how LLMs could elevate your approach to web interaction, moving beyond scraping to true digital engagement.

Green Background

Control All your GenAI Apps in minutes