Every useful LLM application eventually needs data from the web: product docs for a support bot, articles for a research agent, today's prices for a pricing tool. Most tutorials stop at "fetch the page and paste it into the prompt". That works for a demo and breaks in production: you pay for every tag the model reads, menus and cookie banners pollute your embeddings, and sites time out, rate-limit you or hide the real data in JSON.
This book is the missing middle. In short Python programs you can read in one sitting, it shows how to turn any web page into clean, structured, validated input for language models: polite fetching with timeouts, retries and robots.txt; HTML to clean Markdown with measured token savings; structured data you get for free (JSON-LD, Open Graph); schema-first extraction with Pydantic; letting an LLM extract fields and checking its work; chunking for RAG; crawling whole sites; and watching pages for changes.
Published in progress: chapters 1 to 3 are ready now, the free sample includes the preface and chapters 1 and 2, and every buyer gets every new chapter for free as it is released.
