LLMs are trained on massive web datasets like Common Crawl, but this leads to knowledge cutoffs and hallucinations.
Retrieval-Augmented Generation (RAG), or context engineering, feeds external, live data to LLMs to produce more accurate and timely answers.
Modern websites use dynamic JavaScript rendering and anti-bot measures, requiring headless browsers, proxies, and CAPTCHA solvers to access data.
To avoid the 'garbage in, garbage out' problem, you must clean HTML by removing cookie banners and ads, and manage complexities like sitemaps and robots.txt.
A demonstration shows how to use the Apify Website Content Crawler to perform a deep crawl of a website and extract its content into markdown.
The scraped website data is uploaded to a Pinecone vector database, enabling a chatbot to answer questions using the site's specific content.
The Model Context Protocol (MCP) provides a fluid, dynamic interface for AI agents to communicate with and discover tools, unlike static traditional APIs.
An AI agent uses MCP to dynamically search the Apify store for a Twitter scraper, add it to its context, and then use it to fetch live data.
26:15Vidas Bacevičius
23:17Angie Jones
21:17Carl Lapierre
25:32Viktoria Semaan
21:17Lars Kölker
28:04Phil Nash
26:25Ashish Sharma
28:20Dieter Flick
Jobs that call for the skills explored in this talk.
Gemma Analytics
Conrad Electronic SE